Papers

We propose a novel zero-shot multi-frame image restoration method for removing unwanted obstruction elements (such as rains, snow, and moire patterns) that vary in successive frames. It has three stages: transformer pre-training, zero-shot restoration, and hard patch refinement. Using the pre-trained transformers, our model is able to tell the motion difference between the true image information and the obstructing elements. For zero-shot image restoration, we design a novel model, termed SiamTrans, which is constructed by Siamese transformers, encoders, and decoders. Each transformer has a temporal attention layer and several self-attention layers, to capture both temporal and spatial information of multiple frames. Only unsupervisedly pre-trained on the denoising task, SiamTrans is tested on three different low-level vision tasks (deraining, demoireing, and desnowing). Compared with related methods, SiamTrans achieves the best performances, even outperforming those with supervised learning.

poster session 4 @ red 3, poster session 8 @ red 3, oral session 8 @ red 3, poster session 4, poster session 8, oral session 8

How deep neural networks (DNNs) learn from noisy labels has been studied extensively in image classification but much less in image segmentation. So far, our understanding of the learning behavior of DNNs trained by noisy segmentation labels remains limited. In this study, we address this deficiency in both binary segmentation of biological microscopy images and multi-class segmentation of natural images. We generate extremely noisy labels by randomly sampling a small fraction (e.g., 10%) or flipping a large fraction (e.g., 90%) of the ground truth labels. When trained with these noisy labels, DNNs provide largely the same segmentation performance as trained by the original ground truth. This indicates that DNNs learn structures hidden in labels rather than pixel-level labels per se in their supervised training for semantic segmentation. We refer to these hidden structures in labels as meta-structures. When DNNs are trained by labels with different perturbations to the meta-structure, we find consistent degradation in their segmentation performance. In contrast, incorporation of meta-structure information substantially improves performance of an unsupervised segmentation model developed for binary semantic segmentation. We define meta-structures mathematically as spatial density distributions and show both theoretically and experimentally how this formulation explains key observed learning behavior of DNNs.

poster session 4 @ red 3, poster session 8 @ red 3, oral session 8 @ red 3, poster session 4, poster session 8, oral session 8

Clustering is important for domain adaptive person re-identification(re-ID). A majority of unsupervised domain adaptation (UDA) methods conduct clustering on the target domain and then use the generated pseudo labels for adaptive training. Albeit important, the clustering pipeline adopted by current literature is quite standard and lacks consideration for two characteristics of re-ID, i.e., 1) a single person has various feature distribution in multiple cameras. 2) a person’s occurrence in the same camera are usually temporally continuous. We argue that the multi-camera distribution hinders clustering because it enlarges the intra-class distances. In contrast, the temporal continuity prior is beneficial, because it offers clue for distinguishing some look-alike person (who are temporally far away from each other). These two insight motivate us to propose a novel Divide-And-Regroup Clustering (DARC) pipeline for re-ID UDA. Specifically, DARC divides the unlabeled data into multiple camera-specific groups and conducts local clustering within each camera. Afterwards, it regroups those local clusters potentially belonging to the same person into a unity. Through this divide-and-regroup pipeline, DARC avoids directly clustering across multiple cameras and focuses on the feature distribution within each individual camera. Moreover, during the local clustering, DARC uses the temporal continuity prior to distinguish some look-alike person and thus reduces false positive pseudo labels. Consequentially, DARC effectively reduces clustering errors and improves UDA. Importantly, we show that DARC is compatible to many pseudo label-based UDA methods and brings general improvement. Based on a recent UDA method, DARC advances the state of the art (e.g, 85.1% mAP on MSMT-to-Market and 83.1% mAP on PersonX-to-Market).

poster session 3 @ red 1, poster session 8 @ red 1, poster session 3, poster session 8

Despite the many recent practical and theoretical breakthroughs in computational game theory, equilibrium finding in extensive-form team games remains a significant challenge. While NP-hard in the worst case, there are provably efficient algorithms for certain families of team game. In particular, if the game has common external information, also known as A-loss recall---informally, actions played by non-team members (i.e., the opposing team or nature) are either unknown to the entire team, or common knowledge within the team---then polynomial-time algorithms exist (Kaneko and Kline 1995). In this paper, we devise a completely new algorithm for solving team games. It uses a tree decomposition of the constraint system representing each team's strategy to reduce the number and degree of constraints required for correctness (tightness of the mathematical program). Our approach has the bags of the tree decomposition correspond to team-public states---that is, minimal sets of nodes (that is, states of the team) such that, upon reaching the set, it is common knowledge among the players on the team that the set has been reached. Our algorithm reduces the problem of solving team games to a linear program with at most O(NW^{w+1}) nonzero entries in the constraint matrix, where N is the size of the game tree, w is a parameter that depends on the amount of uncommon external information, and W is the treewidth of the tree decomposition. In public-action games, our program size is bounded by the tighter 2^{O(nt)}N for teams of n players with t types each. Our algorithm is based on a new way to write a custom, concise tree decomposition, and its fast run time does not assume that the decomposition has small treewidth. Since our algorithm describes the polytope of correlated strategies directly, we get equilibrium finding in correlated strategies for free---instead of, say, having to run a double oracle algorithm. We show via experiments on a standard suite of games that our algorithm achieves state-of-the-art performance on all benchmark game classes except one. We also present, to our knowledge, the first experiments for this setting where both teams have more than one member.

poster session 5 @ blue 6, poster session 8 @ blue 6, poster session 5, poster session 8

When designing algorithms for finite-time-horizon episodic reinforcement learning problems, a common approach is to introduce a fictitious discount factor and use stationary policies for approximations. Empirically, it has been shown that the fictitious discount factor helps reduce variance, and stationary policies serve to save the per-iteration computational cost. Theoretically, however, there is no existing work on convergence analysis for algorithms with this fictitious discount recipe. This paper takes the first step towards analyzing these algorithms. It focuses on two vanilla policy gradient (VPG) variants: the first being a widely used variant with discounted advantage estimations (DAE), the second with an additional fictitious discount factor in the score functions of the policy gradient estimators. Non-asymptotic convergence guarantees are established for both algorithms, and the additional discount factor is shown to reduce the bias introduced in DAE and thus improve the algorithm convergence asymptotically. A key ingredient of our analysis is to connect three settings of Markov decision processes (MDPs): the finite-time-horizon, the average reward and the discounted settings. To our best knowledge, this is the first theoretical guarantee on fictitious discount algorithms for the episodic reinforcement learning of finite-time-horizon MDPs, which also leads to the (first) global convergence of policy gradient methods for finite-time-horizon episodic reinforcement learning.

poster session 1 @ blue 1, poster session 8 @ blue 1, poster session 1, poster session 8

Recovering detailed facial geometry from a set of calibrated multi-view images is valuable for its wide range of applications. Traditional multi-view stereo (MVS) methods adopt optimization methods to regularize the matching cost. Recently, learning-based methods integrate all these into an end-to-end neural network and show superiority of efficiency. In this paper, we propose a novel architecture to recover extremely detailed 3D faces in roughly 10 seconds. Unlike previous learning-based methods that regularize the cost volume via 3D CNN, we propose to learn an implicit function for regressing the matching cost. By fitting a 3D morphable model from multi-view images, the features of multiple images are extracted and aggregated in the mesh-attached UV space, which makes the implicit function more effective in recovering detailed facial shape. Our method outperforms SOTA learning-based MVS in accuracy by a large margin on the FaceScape dataset. The code and data are released in https://github.com/zhuhao-nju/mvfr.

poster session 3 @ red 2, poster session 8 @ red 2, oral session 8 @ red 2, poster session 3, poster session 8, oral session 8

Scene graph in a video conveys a wealth of information about objects and their relationships in the scene, thus benefiting many downstream tasks such as video captioning and visual question answering. Existing methods of scene graph generation require large-scale training videos annotated with objects and relationships in each frame to learn a powerful model. However, such comprehensive annotation is time-consuming and labor-intensive. On the other hand, it is much easier and less cost to annotate images with scene graphs, so we investigate leveraging annotated images to facilitate training a scene graph generation model for unannotated videos, namely image-to-video scene graph generation. This task presents two challenges: 1) infer unseen dynamic relationships in videos from static relationships in images due to the absence of motion information in images; 2) adapt objects and static relationships from images to video frames due to the domain shift between them. To address the first challenge, we exploit external commonsense knowledge to infer the unseen dynamic relationship from the temporal evolution of static relationships. We tackle the second challenge by hierarchical adversarial learning to reduce the data distribution discrepancy between images and video frames. Extensive experiment results on two benchmark video datasets demonstrate the effectiveness of our method.

poster session 3 @ blue 3, poster session 8 @ blue 3, oral session 3 @ blue 3, poster session 3, poster session 8, oral session 3

Category-level 6D pose estimation can be better generalized to unseen objects in a category compared with instancelevel 6D pose estimation. However, existing category-level 6D pose estimation methods usually require supervised training with a sufficient number of 6D pose annotations of objects which makes them difficult to be applied in real scenarios. To address this problem, we propose a self-supervised framework for category-level 6D pose estimation in this paper.We leverage DeepSDF as a 3D object representation and design several novel loss functions based on DeepSDF to help the self-supervised model predict unseen object poses without any 6D object pose labels and explicit 3D models in real scenarios.Experiments demonstrate that our method achieves comparable performance with the state-of-the-art fully supervised methods on the category-level NOCS benchmark.

poster session 3 @ red 2, poster session 8 @ red 2, poster session 3, poster session 8

Recent years have seen significant advances in multi-turn Spoken Language Understanding (SLU), where dialogue contexts are used to guide intent classification and slot filling. However, how to selectively incorporate dialogue contexts, such as previous utterances and dialogue acts, into multi-turn SLU still remains a substantial challenge. In this work, we propose a novel contextual SLU model for multi-turn intent classification and slot filling tasks. We introduce an adaptive global-local context fusion mechanism to selectively integrate dialogue contexts into our model. The local context fusion aligns each dialogue context using multi-head attention, while the global context fusion measures overall context contribution to intent classification and slot filling tasks. Experiments show that on two benchmark datasets, our model achieves absolute F1 score improvements of 2.73% and 2.57% for the slot filling task on Sim-R and Sim M datasets, respectively. Additional experiments on a large-scale, de-identified, in-house dataset further verify the measurable accuracy gains of our proposed model.

poster session 7 @ red 6, poster session 8 @ red 6, poster session 7, poster session 8

Invariant risk minimization (IRM) has recently emerged as a promising alternative for domain generalization. Nevertheless, the loss function is difficult to optimize for nonlinear classifiers and the original optimization objective could fail when pseudo-invariant features and geometric skews exist. Inspired by IRM, in this paper we propose a novel formulation for domain generalization, dubbed invariant information bottleneck (IIB). IIB aims at minimizing invariant risks for nonlinear classifiers and simultaneously mitigating the impact of pseudo-invariant features and geometric skews. Specifically, we first present a novel formulation for invariant causal prediction via mutual information. Then we adopt the variational formulation of the mutual information to develop a tractable loss function for nonlinear classifiers. To overcome the failure modes of IRM, we propose to minimize the mutual information between the inputs and the corresponding representations. IIB significantly outperforms IRM on synthetic datasets, where the pseudo-invariant features and geometric skews occur, showing the effectiveness of proposed formulation in overcoming failure modes of IRM. Furthermore, experiments on DomainBed show that IIB outperforms $13$ baselines by $0.9\%$ on average across $7$ real datasets.

poster session 4 @ red 3, poster session 8 @ red 3, poster session 4, poster session 8

Neural sequence models trained with maximum likelihood estimation have led to breakthroughs in many tasks, where success is defined by the gap between training and test performance. However, their ability to achieve stronger forms of generalization remains unclear. We consider the problem of symbolic mathematical integration, as it requires generalizing systematically beyond the training set. We develop a methodology for evaluating generalization that takes advantage of the problem domain's structure and access to a verifier. Despite promising in-distribution performance of sequence-to-sequence models in this domain, we demonstrate challenges in achieving robustness, compositionality, and out-of-distribution generalization, through both carefully constructed manual test suites and a genetic algorithm that automatically finds large collections of failures in a controllable manner. Our investigation highlights the difficulty of generalizing well with the predominant modeling and learning approach, and the importance of evaluating beyond the test set, across different aspects of generalization.

poster session 1 @ red 4, poster session 8 @ red 4, poster session 1, poster session 8

Transformer-based pre-trained language models like BERT and its variants have recently achieved promising performance in various natural language processing (NLP) tasks. However, the conventional paradigm constructs the backbone by purely stacking the manually designed global self-attention layers, introducing inductive bias and thus leads to sub-optimal. In this work, we make the first attempt to automatically discover novel pre-trained language model (PLM) backbone on a flexible search space containing the most fundamental operations from scratch. Specifically, we propose a well-designed search space which (i) contains primitive math operations in the intra-layer level to explore novel attention structures, and (ii) leverages convolution blocks to be the supplementary for attentions in the inter-layer level to better learn local dependency. To enhance the efficiency for finding promising architectures, we propose an Operation-Priority Neural Architecture Search (OP-NAS) algorithm, which optimizes both the search algorithm and evaluation of candidate models. Specifically, we propose Operation-Priority (OP) evolution strategy to facilitate model search via balancing exploration and exploitation. Furthermore, we design a Bi-branch Weight-Sharing (BIWS) training strategy for fast model evaluation. Extensive experiments show that the searched architecture (named AutoBERT-Zero) significantly outperforms BERT and its variants of different model capacities in various downstream tasks, proving the architecture's transfer and scaling abilities. Remarkably, AutoBERT-Zero-base outperforms RoBERTa-base (using much more data) and BERT-large (with much larger model size) by 2.4 and 1.4 higher score on GLUE test set.

poster session 3 @ red 5, poster session 8 @ red 5, poster session 3, poster session 8

We introduce a new statistical verification algorithm that formally quantifies the behavioral robustness of any time-continuous process formulated as a continuous-depth model. Our algorithm solves a set of global optimization (Go) problems over a given time horizon to construct a tight enclosure (Tube) of the set of all process executions starting from a ball of initial states. We call our algorithm GoTube. Through its construction, GoTube ensures that the bounding tube is conservative up to a desired probability and up to a desired tightness. GoTube is implemented in JAX and optimized to scale to complex continuous-depth neural network models. Compared to advanced reachability analysis tools for time-continuous neural networks, GoTube does not accumulate overapproximation errors between time steps and avoids the infamous wrapping effect inherent in symbolic techniques. We show that GoTube substantially outperforms state-of-the-art verification tools in terms of the size of the initial ball, speed, time-horizon, task completion, and scalability on a large set of experiments. GoTube is stable and sets the state-of-the-art in terms of its ability to scale to time horizons well beyond what has been previously possible.

poster session 4 @ blue 5, poster session 8 @ blue 5, oral session 4 @ blue 5, poster session 4, poster session 8, oral session 4

Shared autonomy refers to approaches for enabling an autonomous agent to collaborate with a human with the aim of improving human performance. However, besides improving performance, it may often also be beneficial that the agent concurrently accounts for preserving the user’s experience or satisfaction of collaboration. In order to address this additional goal, we examine approaches for improving the user experience by constraining the number of interventions by the autonomous agent. We propose two model-free reinforcement learning methods that can account for both hard and soft constraints on the number of interventions. We show that not only does our method outperform the existing baseline, but also eliminates the need to manually tune a black-box hyperparameter for controlling the level of assistance. We also provide an in-depth analysis of intervention scenarios in order to further illuminate system understanding.

poster session 4 @ blue 5, poster session 8 @ blue 5, poster session 4, poster session 8

We study the problem of safe offline reinforcement learning (RL), the goal is to learn a policy that maximizes long-term reward while satisfying safety constraints given only offline data, without further interaction with the environment. This problem is more appealing for real world RL applications, in which data collection is costly or dangerous. Enforcing constraint satisfaction is non-trivial, especially in offline settings, as there is a potential large discrepancy between the policy distribution and the data distribution, causing errors in estimating the value of safety constraints. We show that na\"ive approaches that combine techniques from safe RL and offline RL can only learn sub-optimal solutions. We thus develop a simple yet effective algorithm, Constraints Penalized Q-Learning (CPQ), to solve the problem. Our method admits the use of data generated by mixed behavior policies. We present a theoretical analysis and demonstrate empirically that our approach can learn robustly across a variety of benchmark control tasks, outperforming several baselines.

poster session 1 @ blue 1, poster session 8 @ blue 1, poster session 1, poster session 8

Motivated by fair division applications, we study a fair connected graph partitioning problem, in which an undirected graph with m nodes must be divided between n agents such that each agent receives a connected subgraph and the partition is fair. We study approximate versions of two fairness criteria: \alpha-proportionality requires that each agent receive a subgraph with at least (1/\alpha)*m/n nodes, and \alpha-balancedness requires that the ratio between the sizes of the largest and smallest subgraphs be at most \alpha. Unfortunately, there exist simple examples in which no partition is reasonably proportional or balanced. To circumvent this, we introduce the idea of charity. We show that by "donating" just n-1 nodes, we can guarantee the existence of 2-proportional and almost 2-balanced partitions (and find them in polynomial time), and that this result is almost tight. More generally, we chart the tradeoff between the size of charity and the approximation of proportionality or balancedness we can guarantee.

poster session 5 @ blue 6, poster session 8 @ blue 6, oral session 8 @ blue 6, poster session 5, poster session 8, oral session 8

The shortcomings of maximum likelihood estimation in the context of model-based reinforcement learning have been highlighted by an increasing number of papers. When the model class is misspecified or has a limited representational capacity, model parameters with high likelihood might not necessarily result in high performance of the agent on a downstream control task. To alleviate this problem, we propose an end-to-end approach for model learning which directly optimizes the expected returns using implicit differentiation. We treat a value function that satisfies the Bellman optimality operator induced by the model as an implicit function of model parameters and show how to differentiate the function. We provide theoretical and empirical evidence highlighting the benefits of our approach in the model misspecification regime compared to likelihood-based methods.

poster session 1 @ blue 1, poster session 8 @ blue 1, poster session 1, poster session 8

Depth estimation is a crucial step for 3D reconstruction with panorama images in recent years. Panorama images maintain the complete spatial information but introduce distortion with equirectangular projection. In this paper, we propose an ACDNet based on the adaptively combined dilated convolution to predict the dense depth map for a monocular panoramic image. Specifically, we combine the convolution kernels with different dilations to extend the receptive field in the equirectangular projection. Meanwhile, we introduce an adaptive channel-wise fusion module to summarize the feature maps and get diverse attention areas in the receptive field along the channels. Due to the utilization of channel-wise attention in constructing the adaptive channel-wise fusion module, the network can capture and leverage the cross-channel contextual information efficiently. Finally, we conduct depth estimation experiments on three datasets (both virtual and real-world) and the experimental results demonstrate that our proposed ACDNet substantially outperforms the current state-of-the-art (SOTA) methods.

poster session 3 @ red 2, poster session 8 @ red 2, poster session 3, poster session 8

Capturing visual similarity among images is the core of many computer vision and pattern recognition tasks. This problem can be formulated in such a paradigm called metric learning. Most research in the area has been mainly focusing on improving the loss functions and similarity measures. However, due to the ignoring of geometric structure, existing methods often lead to sub-optimal results. Thus, several recent research methods took advantage of Wasserstein distance between batches of samples to characterize the spacial geometry. Although these approaches can achieve enhanced performance, the aggregation over batches definitely hinders Wasserstein distance's superior measure capability and leads to high computational complexity. To address this limitation, we propose a novel Deep Wasserstein Metric Learning framework, which employs Wasserstein distance to precisely capture the relationship among various images under ranking-based loss functions such as contrastive loss and triplet loss. Our method directly computes the distance between images, considering the geometry at a finer granularity than batch level. Furthermore, we introduce a new efficient algorithm using Sinkhorn approximation and Wasserstein measure coreset. The experimental results demonstrate the improvements of our framework over various baselines in different applications and benchmark datasets.

poster session 3 @ red 5, poster session 8 @ red 5, poster session 3, poster session 8

We consider a variant of the standard Bayesian mechanism, where players evaluate their outcomes and constraints in an {\em ex-ante} manner. Such a model captures a major form of modern online advertising where an advertiser is concerned with her/his expected utility over a time period and her/his type may change over time. We are interested in the incentive compatibility (IC) problem of such Bayesian mechanism. Under very mild conditions on the mechanism environments, we give a full characterization of IC via the {\em taxation principle} and show, perhaps surprisingly, that such IC mechanisms are fully characterized by the so-called {\em auto-bidding} mechanisms, which are pervasively fielded in the online advertising industry.

poster session 5 @ blue 6, poster session 8 @ blue 6, poster session 5, poster session 8

Detecting 3D lanes from the camera is a rising problem for autonomous vehicles. In this task, the correct camera pose is the key to generating accurate lanes, which can transform an image from perspective-view to the top-view. With this transformation, we can get rid of the perspective effects so that 3D lanes would look similar and can accurately be fitted by low-order polynomials. However, mainstream 3D lane detectors rely on perfect camera poses provided by other sensors, which is expensive and encounters multi-sensor calibration issues. To overcome this problem, we propose to predict 3D lanes by estimating camera pose from a single image with a two-stage framework. The first stage aims at the camera pose task from perspective-view images. To improve pose estimation, we introduce an auxiliary 3D lane task and geometry constraints to benefit from multi-task learning, which enhances consistencies between 3D and 2D, as well as compatibility in the above two tasks. The second stage targets the 3D lane task. It uses previously estimated pose to generate top-view images containing distance-invariant lane appearances for predicting accurate 3D lanes. Experiments demonstrate that, without ground truth camera pose, our method outperforms the state-of-the-art perfect-camera-pose-based methods and has the fewest parameters and computations. Codes are available at https://github.com/liuruijin17/CLGo.

poster session 3 @ red 2, poster session 8 @ red 2, poster session 3, poster session 8

Label-specific features serve as an effective strategy to learn from multi-label data with tailored features accounting for the distinct discriminative properties of each class label. Existing prototype-based label-specific feature transformation approaches work in a three-stage framework, where prototype acquisition, label-specific feature generation and classification model induction are performed independently. Intuitively, this separate framework is suboptimal due to its decoupling nature. In this paper, we make a first attempt towards a unified framework for prototype-based label-specific feature transformation, where the prototypes and the label-specific features are directly optimized for classification. To instantiate it, we propose modelling the prototypes probabilistically by the normalizing flows, which possess adaptive prototypical complexity to fully capture the underlying properties of each class label and allow for scalable stochastic optimization. Then, a label correlation regularized probabilistic latent metric space is constructed via jointly learning the prototypes and the metric-based label-specific features for classification. Comprehensive experiments on 14 benchmark data sets show that our approach outperforms the state-of-the-art counterparts.

poster session 3 @ blue 2, poster session 8 @ blue 2, poster session 3, poster session 8

Language models (LMs) can reproduce (or amplify) toxic language seen during training, which poses a risk to their practical application. In this paper, we conduct extensive experiments to study this phenomenon. We analyze the impact of prompts, decoding strategies and training corpora on the output toxicity. Based on our findings, we propose a simple yet effective unsupervised method for language models to ``detoxify'' themselves without an additional large corpus or external discriminator. Compared to a supervised baseline, our proposed method shows better toxicity reduction with good generation quality in the generated content under multiple settings.

poster session 1 @ red 4, poster session 8 @ red 4, poster session 1, poster session 8

Goal-oriented dialog policy learning algorithms aim to learn a dialog policy for selecting language actions based on the current dialog state. Deep reinforcement learning methods have been used for dialog policy learning. This work is motivated by the observation that, although dialog is a domain with rich contextual knowledge, reinforcement learning methods are ill-equipped to incorporate such knowledge into the dialog policy learning process. In this paper, we develop a deep reinforcement learning framework for goal-oriented dialog policy learning that learns user preferences from user goal data, while leveraging commonsense knowledge from people. The developed framework has been evaluated using a realistic dialog simulation platform. Compared with baselines from the literature and the ablations of our approach, we see significant improvements in learning efficiency and the quality of the computed action policies.

poster session 3 @ red 5, poster session 8 @ red 5, poster session 3, poster session 8

The unit selection problem aims to identify a set of individuals who are most likely to exhibit a desired mode of behavior, for example, selecting individuals who would respond one way if encouraged and a different way if not encouraged. Using a combination of experimental and observational data, Li and Pearl derived tight bounds on the "benefit function" - the payoff/cost associated with selecting an individual with given characteristics. This paper shows that these bounds can be narrowed significantly (enough to change decisions) when structural information is available in the form of a causal model. We address the problem of estimating the benefit function using observational and experimental data when specific graphical criteria are assumed to hold.

poster session 4 @ blue 4, poster session 8 @ blue 4, poster session 4, poster session 8

We propose a theoretical framework of multi-way similarity to model real-valued data into hypergraphs for clustering via spectral embedding. For graph cut based spectral clustering, it is common to model real-valued data into graph by modeling pairwise similarities using kernel function. This is because the kernel function has a theoretical connection to the graph cut. For problems where using multi-way similarities are more suitable than pairwise ones, it is natural to model as a hypergraph, which is generalization of a graph. However, although the hypergraph cut is well-studied, there is not yet established a hypergraph cut based framework to model multi-way similarity. In this paper, we formulate multi-way similarities by exploiting the theoretical foundation of kernel function. We show a theoretical connection between our formulation and hypergraph cut in two ways, generalizing both weighted kernel $k$-means and the heat kernel, by which we justify our formulation. We also provide a fast algorithm for spectral clustering. Our algorithm empirically shows better performance than existing graph and other heuristic modeling methods.

poster session 3 @ blue 2, poster session 8 @ blue 2, poster session 3, poster session 8

Automated persuasion systems (APS) aim to persuade a user to believe something by entering into a dialogue in which arguments and counterarguments are exchanged. To maximize the probability that an APS is successful in persuading a user, it can identify a global policy that will allow it to select the best arguments it presents at each stage of the dialogue whatever arguments the user presents. However, in real applications, such as for healthcare, it is unlikely the utility of the outcome of the dialogue will be the same, or the exact opposite, for the APS and user. In order to deal with this situation, games in extended form have been harnessed for argumentation in Bi-party Decision Theory. This opens new problems that we address in this paper: (1) How can we use Machine Learning (ML) methods to predict utility functions for different subpopulations of users? and (2) How can we identify for a new user the best utility function from amongst those that we have learned. To this extent, we develop two ML methods, EAI and EDS, that leverage information coming from the users to predict their utilities. EAI is restricted to a fixed amount of information, whereas EDS can choose the information that best detects the subpopulations of a user. We evaluate EAI and EDS in a simulation setting and in a realistic case study concerning healthy eating habits. Results are promising in both cases, but EDS is more effective at predicting useful utility functions.

poster session 4 @ blue 5, poster session 8 @ blue 5, oral session 4 @ blue 5, poster session 4, poster session 8, oral session 4

3D object detection from point clouds data has become an indispensable part in autonomous driving. Previous works for processing point clouds lie in either projection or voxelization. However, projection-based methods suffer from information loss while voxelization-based methods bring huge computation. In this paper, we propose to encode point clouds into structured color image representation (SCIR) and utilize 2D CNN to fulfill the 3D detection task. Specifically, we use the structured color image encoding module to convert the irregular 3D point clouds into a squared 2D tensor image, where each point corresponds to a spatial point in the 3D space. Furthermore, in order to fit for the Euclidean structure, we apply feature normalization to parameterize the 2D tensor image onto a regular dense color image. Then, we conduct repeated multi-scale fusion with different levels so as to augment the initial features and learn scale-aware feature representations for box prediction. Extensive experiments on KITTI benchmark, Waymo Open Dataset and more challenging nuScenes dataset show that our proposed method yields decent results and demonstrate the effectiveness of such representations for point clouds.

poster session 3 @ red 2, poster session 8 @ red 2, poster session 3, poster session 8

We posit a new mechanism for cooperation in multi-agent reinforcement learning (MARL) based upon any nonlinear function of the team's long-term state-action occupancy measure, i.e., a \emph{general utility}. This subsumes the cumulative return but also allows one to incorporate risk-sensitivity, exploration, and priors. % We derive the {\bf D}ecentralized {\bf S}hadow Reward {\bf A}ctor-{\bf C}ritic (DSAC) in which agents alternate between policy evaluation (critic), weighted averaging with neighbors (information mixing), and local gradient updates for their policy parameters (actor). DSAC augments the classic critic step by requiring agents to (i) estimate their local occupancy measure in order to (ii) estimate the derivative of the local utility with respect to their occupancy measure, i.e., the ``shadow reward". DSAC converges to $\epsilon$-stationarity in $\mathcal{O}(1/\epsilon^{2.5})$ (Theorem \ref{theorem:final}) or faster $\mathcal{O}(1/\epsilon^{2})$ (Corollary \ref{corollary:communication}) steps with high probability, depending on the amount of communications. We further establish the non-existence of spurious stationary points for this problem, that is, DSAC finds the globally optimal policy (Corollary \ref{corollary:global}). Experiments demonstrate the merits of goals beyond the cumulative return in cooperative MARL.

poster session 1 @ blue 1, poster session 8 @ blue 1, poster session 1, poster session 8

Pre-trained models have proved to be powerful in enhancing task-oriented dialog systems. However, current pre-training methods mainly focus on enhancing dialog understanding and generation tasks while neglecting the exploitation of dialog policy. In this paper, we propose GALAXY, a novel pre-trained dialog model that explicitly learns dialog policy from limited labeled dialogs and large-scale unlabeled dialog corpora via semi-supervised learning. Specifically, we introduce a dialog act prediction task for policy optimization during pre-training and employ a consistency regularization term to refine the learned representation with the help of unlabeled dialogs. We also implement a gating mechanism to weigh suitable unlabeled dialog samples. Empirical results show that GALAXY substantially improves the performance of task-oriented dialog systems, and achieves new state-of-the-art results on benchmark datasets: In-Car, MultiWOZ2.0 and MultiWOZ2.1, improving their end-to-end combined scores by 2.5, 5.3 and 5.5 points, respectively. We also show that GALAXY has a stronger few-shot ability than existing models under various low-resource settings. For reproducibility, we release the code and data at https://github.com/siat-nlp/GALAXY.

poster session 3 @ red 5, poster session 8 @ red 5, poster session 3, poster session 8

In this paper, we propose a novel problem, named rational (alpha, beta)-core detection in attribute bipartite graphs (RCD-ABG), which retrieves the connected (alpha, beta)-core with the largest rational score. A basic greedy framework with an optimized strategy is developed and extensive experiments are conducted to evaluate the performance of the techniques.

poster session 3 @ blue 2, poster session 8 @ blue 2, poster session 3, poster session 8

Ensemble Learning is an effective method for improving generalization in machine learning. However, as state-of-the-art neural networks grow larger, the computational cost associated with training several independent networks becomes expensive. We introduce a fast, low-cost method for creating diverse ensembles of neural networks without needing to train multiple models from scratch. We do this by first training a single parent network. We then create child networks by cloning the parent and dramatically pruning the parameters of each child to create an ensemble of members with unique and diverse topologies. We then briefly train each child network for a small number of epochs, which now converge significantly faster when compared to training from scratch. We explore various ways to maximize diversity in the child networks, including the use of anti-random pruning and one-cycle tuning. This diversity enables "Prune and Tune" ensembles to achieve results that are competitive with traditional ensembles at a fraction of the training cost. We benchmark our approach against state of the art low-cost ensemble methods and display marked improvement in both accuracy and uncertainty estimation on CIFAR-10 and CIFAR-100.

poster session 4 @ blue 5, poster session 8 @ blue 5, poster session 4, poster session 8

Recent studies have shown that strong Natural Language Understanding (NLU) models are prone to relying on annotation biases of the datasets as a shortcut, which goes against the underlying mechanisms of the task of interest. To reduce such biases, several recent works introduce debiasing methods to regularize the training process of targeted NLU models. In this paper, we provide a new perspective with causal inference to find out the bias. On one hand, we show that there is an unobserved confounder for the natural language utterances and their respective classes, leading to spurious correlations from training data. To remove such confounder, the backdoor adjustment with causal intervention is utilized to find the true causal effect, which makes the training process fundamentally different from the traditional likelihood estimation. On the other hand, in inference process, we formulate the bias as the direct causal effect and remove it by pursuing the indirect causal effect with counterfactual reasoning. We conduct experiments on large-scale natural language inference and fact verification benchmarks, evaluating on bias sensitive datasets that are specifically designed to assess the robustness of models against known biases in the training data. Experimental results show that our proposed debiasing framework outperforms previous state-of-the-art debiasing methods while maintaining the original in-distribution performance.

poster session 3 @ red 5, poster session 8 @ red 5, poster session 3, poster session 8

Controlling neural network-based models for natural language generation (NLG) to realize desirable attributes in the generated outputs has broad applications in numerous areas such as machine translation, document summarization, and dialog systems. Approaches that enable such control in a zero-shot manner would be of great importance as, among other reasons, they remove the need for additional annotated data and training. In this work, we propose novel approaches for controlling encoder-decoder transformer-based NLG models in zero shot. While zero-shot control has previously been observed in massive models (e.g., GPT3), our method enables such control for smaller models. This is done by applying two control knobs, attention biasing and context augmentation, to these models directly during decoding and without additional training or auxiliary models. These knobs control the generation process by directly manipulating trained NLG models (e.g., biasing cross-attention layers). We show that not only are these NLG models robust to such manipulations but also their behavior could be controlled without an impact on their generation performance.

poster session 1 @ red 4, poster session 8 @ red 4, poster session 1, poster session 8

We study a dynamic model of Bayesian persuasion in sequential decision-making settings. An informed principal observes an external parameter of the world and advises an uninformed agent about actions to take over time. The agent takes actions in each time step based on the current state, the principal's advice/signal, and beliefs about the external parameter. The action of the agent updates the state according to a stochastic process. The model arises naturally in many applications, e.g., an app (the principal) can advice the user (the agent) on possible choices between actions based on additional real-time information the app has. We study the problem of designing a signaling strategy from the principal's point of view. We show that the principal has an optimal strategy against a myopic agent, who only optimizes their rewards locally, and the optimal strategy can be computed in polynomial time. In contrast, it is NP-hard to approximate an optimal policy against a far-sighted agent. Further, we show that if the principal has the power to threaten the agent by not providing future signals, then we can efficiently design a threat-based strategy. This strategy guarantees the principal's payoff as if playing against an agent who is far-sighted but myopic to future signals.

poster session 5 @ blue 6, poster session 8 @ blue 6, oral session 5 @ blue 6, poster session 5, poster session 8, oral session 5

Unsupervised graph-level representation learning plays a crucial role in a variety of tasks such as molecular property prediction and community analysis, especially when data annotation is expensive. Currently, most of the best-performing graph embedding methods are based on Infomax principle. The performance of these methods highly depends on the selection of negative samples and hurt the performance, if the samples were not carefully selected. Inter-graph similarity-based methods also suffer if the selected set of graphs for similarity matching is low in quality. To address this, we focus only on utilizing the current input graph for embedding learning. We are motivated by an observation from real-world graph generation processes where the graphs are formed based on one or more global factors which are common to all elements of the graph (e.g., topic of a discussion thread, solubility level of a molecule). We hypothesize extracting these common factors could be highly beneficial. Hence, this work proposes a new principle for unsupervised graph representation learning: Graph-wise Common latent Factor EXtraction (GCFX). We further propose a deep model for GCFX, deepGCFX, based on the idea of reversing the above-mentioned graph generation process which could explicitly extract common latent factors from an input graph and achieve improved results on downstream tasks to the current state-of-the-art. Through extensive experiments and analysis, we demonstrate that, while extracting common latent factors is beneficial for graph-level tasks to alleviate distractions caused by local variations of individual nodes or local neighbourhoods, it also benefits node-level tasks by enabling long-range node dependencies, especially for disassortative graphs.

poster session 1 @ blue 1, poster session 8 @ blue 1, oral session 8 @ blue 1, poster session 1, poster session 8, oral session 8

In recent years, companies in the Architecture, Engineering, and Construction (AEC) industry have started exploring how artificial intelligence (AI) can reduce time-consuming and repetitive tasks. One use case that can benefit from the adoption of AI is the determination of quantities in floor plans. This information is required for several planning and construction steps. Currently, the task requires companies to invest a significant amount of manual effort. Either digital floor plans are not available for existing buildings, or the formats cannot be processed due to lack of standardization. In this paper, we therefore propose a human-in-the-loop approach for the detection and classification of symbols in floor plans. The developed system calculates a measure of uncertainty for each detected symbol which is used to acquire the knowledge of human experts for those symbols that are difficult to classify. We evaluate our approach with a real-world dataset provided by an industry partner and find that the selective acquisition of human expert knowledge enhances the model’s performance by up to 10.5%—resulting in an overall prediction accuracy of 92.1% on average. We further design a pipeline for the generation of synthetic training data that allows the systems to be adapted to new construction projects with minimal manual effort. Overall, our work supports professionals in the AEC industry on their journey to the data-driven generation of business value.

poster session 4 @ red 6, poster session 8 @ red 6, poster session 4, poster session 8

Binary optimization plays an important role in mathematical optimization and has various applications in computer vision and machine learning. Generally speaking, binary optimization problems are NP-hard and difficult to solve due to the binary constraints, especially when the number of variables is very large. Existing methods often suffer from high computational costs or large accumulated quantization errors, or are only designed for specific tasks. In this paper, we propose an efficient algorithm, named Discrete Principal Coordinate Descent (DPCD), to find effective approximate solutions for general binary optimization problems. The proposed algorithm supports a wide range of empirical objective functions with/without restrictions on the numbers of 1s and -1s in the binary variables. Furthermore, the theoretical convergence is proven, and the explicit convergence rates are derived for objective functions with Lipschitz continuous gradients, which are commonly adopted in practice. Extensive experiments on several tasks and large-scale datasets demonstrate the superiority of the proposed algorithm over several state-of-the-art methods in terms of both effectiveness and efficiency.

poster session 3 @ red 1, poster session 8 @ red 1, poster session 3, poster session 8

In federated learning (FL), a server determines a global learning model by aggregating the local learning models of clients, and the determined global model is broadcast to all the clients. However, the global learning model can significantly deteriorate if a Byzantine attacker transmits malicious learning models trained with incorrectly labeled data. We propose a Byzantine-robust FL algorithm that, by employing a consensus confirmation method, can reduce the success probability of Byzantine attacks. After aggregating the local models from clients, the proposed FL server validates the global model candidate by sending the global model candidate to a set of randomly selected FL clients and asking them to perform local validation with their local data. If most of the validation is positive, the global model is confirmed and broadcast to all the clients. We compare the performance of the proposed FL against Byzantine attacks with that of existing FL algorithms analytically and empirically.

poster session 3 @ blue 3, poster session 8 @ blue 3, poster session 3, poster session 8

Graph Convolutional Neural Networks (GCNs) have been generally accepted to be an effective tool for node representations learning. An interesting way to understand GCNs is to think of them as a message passing mechanism where each node updates its representation by accepting information from its neighbours (also known as positive samples). However, beyond these neighbouring nodes, graphs have a large, dark, all-but forgotten world in which we find the non-neighbouring nodes (negative samples). In this paper, we show that this great dark world holds a substantial amount of information that might be useful for representation learning. Most specifically, it can provide negative information about the node representations. Our overall idea is to select appropriate negative samples for each node and incorporate the negative information contained in these samples into the representation updates. Moreover, we show that the process of selecting the negative samples is not trivial. Our theme therefore begins by describing the criteria for a good negative sample, followed by a determinantal point process algorithm for efficiently obtaining such samples. A GCN, boosted by diverse negative samples, then jointly considers the positive and negative information when passing messages. Experimental evaluations show that this idea not only improves the overall performance of standard representation learning but also significantly alleviates over-smoothing problems.

poster session 3 @ blue 2, poster session 8 @ blue 2, poster session 3, poster session 8

In this paper, we investigate the generalization problem of person re-identification (re-id), whose major challenge is the distribution shift on an unseen domain. As an important tool of regularizing the distribution, batch normalization (BN) has been widely used in existing methods. However, they neglect that BN is severely biased to the training domain and inevitably suffers the performance drop if directly generalized without being updated. To tackle this issue, we propose Batch Norm Test-time Adaption (BNTA), a novel re-id framework that applies the self-supervised strategy to update BN parameters adaptively. Specifically, BNTA quickly explores the domain-aware information within unlabeled target data before inference, and accordingly modulates the feature distribution normalized by BN to adapt to the target domain. This is accomplished by two designed self-supervised auxiliary tasks, namely part positioning and part nearest neighbor matching, which help the model mine the domain-aware information with respect to the structure and identity of body parts, respectively. To demonstrate the effectiveness of our method, we conduct extensive experiments on three re-id datasets and confirm the superior performance to the state-of-the-art methods.

poster session 3 @ red 1, poster session 8 @ red 1, poster session 3, poster session 8

Accurate 3D object detection from point clouds has become a crucial component in autonomous driving. However, the volumetric representations and the projection methods in previous works fail to establish the relationships between the local point sets. In this paper, we propose Sparse Voxel-Graph Attention Network (SVGA-Net), a novel end-to-end trainable network which mainly contains voxel-graph module and sparse-to-dense regression module to achieve comparable 3D detection tasks from raw LIDAR data. Specifically, SVGA-Net constructs the local complete graph within each divided 3D spherical voxel and global KNN graph through all voxels. The local and global graphs serve as the attention mechanism to enhance the extracted features. In addition, the novel sparse-to-dense regression module enhances the 3D box estimation accuracy through feature maps aggregation at different levels. Experiments on KITTI detection benchmark and Waymo Open dataset demonstrate the efficiency of extending the graph representation to 3D object detection and the proposed SVGA-Net can achieve decent detection accuracy.

poster session 3 @ red 2, poster session 8 @ red 2, poster session 3, poster session 8

3D meshes are widely employed to represent geometry structure of 3D shapes. Due to limitation of scanning sensor precision and other issues, meshes are inevitably affected by noise, which hampers the subsequent applications. Convolultional neural networks (CNNs) achieve great success in image processing tasks, including 2D image denoising, and have been proven to own the capacity of modeling complex features at different scales, which is also particularly useful for mesh denoising. However, due to the nature of irregular structure, CNNs-based denosing strategies cannot be trivially applied for meshes. To circumvent this limitation, in the paper, we propose the local surface descriptor (LSD), which is able to transform the local deformable surface around a face into 2D grid representation and thus facilitates the deployment of CNNs to generate denoised face normals. To verify the superiority of LSD, we directly feed LSD into the classical Resnet without any complicated network design. The extensive experimental results show that, compared to the state-of-the-arts, our method achieves encouraging performance with respect to both objective and subjective evaluations.

poster session 3 @ red 2, poster session 8 @ red 2, poster session 3, poster session 8

Off-policy learning plays a pivotal role in optimizing and evaluating policies prior to the online deployment. However, during the real-time serving, we observe varieties of interventions and constraints that cause inconsistency between the online and offline setting, which we summarize and term as runtime uncertainty. Such uncertainty cannot be learned from the logged data due to its abnormality and rareness nature. To assert a certain level of robustness, we perturb the off-policy estimators along an adversarial direction in view of the runtime uncertainty. It allows the resulting estimators to be robust not only to observed but also unexpected runtime uncertainties. Leveraging this idea, we bring runtime-uncertainty robustness to three major off-policy learning methods: the inverse propensity score method, reward-model method, and doubly robust method. We theoretically justify the robustness of our methods to runtime uncertainty, and demonstrate their effectiveness using both the simulation and the real-world online experiments.

poster session 4 @ blue 4, poster session 8 @ blue 4, poster session 4, poster session 8

The paper presents a characterization of logic programs with aggregates based on many-sorted generalization of operator SM that refers neither to grounding nor to fixpoints. This characterization introduces new symbols for aggregate operations and aggregate elements, whose meaning is fixed by adding appropriate axioms to the result of the SM transformation. We prove that for programs without positive recursion through aggregates our semantics coincides with the semantics of the answer set solver Clingo.

poster session 4 @ blue 5, poster session 8 @ blue 5, oral session 8 @ blue 5, poster session 4, poster session 8, oral session 8

Unsupervised point cloud registration algorithm usually suffers from the unsatisfied registration precision in the partially overlapping problem due to the lack of effective inlier evaluation. In this paper, we propose a neighborhood consensus based reliable inlier evaluation method for robust unsupervised point cloud registration. It is expected to capture the discriminative geometric difference between the source neighborhood and the corresponding pseudo target neighborhood for effective inlier distinction. Specifically, our model consists of a matching map refinement module and an inlier evaluation module. In our matching map refinement module, we improve the point-wise matching map estimation by integrating the matching scores of neighbors into it. The aggregated neighborhood information potentially facilitates the discriminative map construction so that high-quality correspondences can be provided for generating the pseudo target point cloud. Based on the observation that the outlier has the significant structure-wise difference between its source neighborhood and corresponding pseudo target neighborhood while this difference for inlier is small, the inlier evaluation module exploits this difference to score the inlier confidence for each estimated correspondence. In particular, we construct an effective graph representation for capturing this geometric difference between the neighborhoods. Finally, with the learned correspondences and the corresponding inlier confidence, we use the weighted SVD algorithm for transformation estimation.Under the unsupervised setting, we exploit the Huber function based global alignment loss, the local neighborhood consensus loss and spatial consistency loss for model optimization. The experimental results on extensive datasets demonstrate that our unsupervised point cloud registration method can yield comparable performance.

poster session 3 @ red 2, poster session 8 @ red 2, poster session 3, poster session 8

Recently, deep learning techniques are soaring and have shown dramatic improvements in real-world noisy image denoising. However, the statistics of real noise generally vary with different camera sensors and in-camera signal processing pipelines. This will induce problems of most deep denoisers for the overfitting or degrading performance due to the noise discrepancy between the training and test sets. To remedy this issue, we propose a novel flexible and adaptive denoising network, coined as FADNet. Our FADNet is equipped with a plane dynamic filter module, which generates weight filters with flexibility that can adapt to the specific input and thereby impedes the FADNet from overfitting to the training data. Specifically, we exploit the advantage of the spatial and channel attention, and utilize this to devise a decoupling filter generation scheme. The generated filters are conditioned on the input and collaboratively applied to the decoded features for representation capability enhancement. We additionally introduce the Fourier transform and its inverse to guide the predicted weight filters to adapt to the noisy input with respect to the image contents. Experimental results demonstrate the superior denoising performances of the proposed FADNet versus the state-of-the-art. In contrast to the existing deep denoisers, our FADNet is not only flexible and efficient, but also exhibits a compelling generalization capability, enjoying tremendous potential for practical usage.

poster session 3 @ red 1, poster session 8 @ red 1, poster session 3, poster session 8

The Linear Threshold Model is a widely used model that describes how information diffuses through a social network. According to this model, an individual adopts an idea or product after the proportion of their neighbors who have adopted it reaches a certain threshold. Typical applications of the Linear Threshold Model assume that thresholds are either the same for all network nodes or randomly distributed, even though some people may be more susceptible to peer pressure than others. To address individual-level differences, we propose causal inference methods for estimating individual thresholds that can more accurately predict whether and when individuals will be affected by their peers. We introduce the concept of heterogeneous peer effects and develop a Structural Causal Model which corresponds to the Linear Threshold Model and supports heterogeneous peer effect identification and estimation. We develop two algorithms for individual threshold estimation, one based on causal trees and one based on causal meta-learners. Our experimental results on synthetic and real- world datasets show that our proposed models can better predict individual-level thresholds in the Linear Threshold Model and thus more precisely predict which nodes will get activated over time.

poster session 4 @ blue 4, poster session 8 @ blue 4, poster session 4, poster session 8

We propose a dual-domain generative model to estimate a texture map from a single image for colorizing a 3D human model. When estimating a texture map, a single image is insufficient as it reveals only one facet of a 3D object. To provide sufficient information for estimating a complete texture map, the proposed model simultaneously generates multi-view hallucinations in the image domain and an estimated texture map in the texture domain. During the generating process, each domain generator exchanges features to the other by a flow-based local attention mechanism. In this manner, the proposed model can estimate a texture map utilizing abundant multi-view image features from which multi-view hallucinations are generated. As a result, the estimated texture map contains consistent colors and patterns over the entire region. Experiments show the superiority of our model for estimating a directly render-able texture map, which is applicable to 3D animation rendering. Furthermore, our model also improves an overall generation quality in the image domain for pose and viewpoint transfer tasks.

poster session 3 @ red 1, poster session 8 @ red 1, poster session 3, poster session 8

When crowdsourced workers perform annotation tasks in an unfamiliar domain, their accuracy will dramatically decline due to the lack of expertise. Transferring knowledge from relevant domains can form a better representation for instances, which benefits the estimation of workers' expertise in truth inference models. However, existing knowledge transfer processes for crowdsourcing require a considerable number of well-collected instances in source domains. This paper proposes a novel truth inference model for crowdsourcing, where (meta-)knowledge is transferred by meta-learning and used in the estimation of workers' expertise. Our preliminary experiments demonstrate that the meta-knowledge transfer significantly reduces instances in source domains and increases the accuracy of truth inference.

poster session 3 @ blue 3, poster session 8 @ blue 3, poster session 3, poster session 8

Obtaining a well-trained model involves expensive data collection and training procedures, therefore the model is a valuable intellectual property. Recent studies revealed that adversaries can `steal' deployed models even when they have no training samples and can not get access to the model parameters or structures. Currently, there were some defense methods to alleviate this threat, mostly by increasing the cost of model stealing. In this paper, we explore the defense from another angle by verifying whether a suspicious model contains the knowledge of defender-specified \emph{external features}. Specifically, we embed the external features by tempering a few training samples with style transfer. We then train a meta-classifier to determine whether a model is stolen from the victim. This approach is inspired by the understanding that the stolen models should contain the knowledge of features learned by the victim model. We examine our method on both CIFAR-10 and ImageNet datasets. Experimental results demonstrate that our method is effective in detecting different types of model stealing simultaneously, even if the stolen model is obtained via a multi-stage stealing process. The codes for reproducing main results are available at Github (https://github.com/zlh-thu/StealingVerification).

poster session 4 @ red 3, poster session 8 @ red 3, poster session 4, poster session 8

Key information extraction (KIE) from document images requires understanding the contextual and spatial semantics of texts in two-dimensional (2D) space. Many recent studies try to solve the task by developing pre-trained language models focusing on combining visual features from document images with texts and their layout. On the other hand, this paper tackles the problem by going back to the basic: effective combination of text and layout. Specifically, we propose a pre-trained language model, named BROS (BERT Relying On Spatiality), that encodes relative positions of texts in 2D space and learns from unlabeled documents with area-masking strategy. With this optimized training scheme for understanding texts in 2D space, BROS shows comparable or better performance compared to previous methods on four KIE benchmarks (FUNSD, SROIE*, CORD, and SciTSR) without relying on visual features. This paper also reveals two real-world challenges in KIE tasks-(1) minimizing the error from incorrect text ordering and (2) efficient learning from fewer downstream examples-and demonstrates the superiority of BROS over previous methods. Code will be available at https://github.com/clovaai.

poster session 3 @ red 5, poster session 8 @ red 5, poster session 3, poster session 8

Prediction using the ground truth sounds like an oxymoron in machine learning. However, such an unrealistic setting was used in hundreds, if not thousands of papers in the area of finding graph representations. To evaluate the multi-label problem of node classification by using the obtained representations, many works assume in the prediction stage that the number of labels of each test instance is known. In practice such ground truth information is rarely available, but we point out that such an inappropriate setting is now ubiquitous in this research area. We detailedly investigate why the situation occurs. Our analysis indicates that with unrealistic information, the performance is likely over-estimated. To see why suitable predictions were not used, we identify difficulties in applying some multi-label techniques. For the use in future studies, we propose simple and effective settings without using practically unknown information. Finally, we take this chance to conduct a fair and serious comparison of major graph-representation learning methods on multi-label node classification.

poster session 3 @ blue 2, poster session 8 @ blue 2, poster session 3, poster session 8

We extend the classic regret minimization framework for approximating equilibria in matrix games by greedily weighing iterates based on the regrets observed at runtime. Theoretically, our method retains all previous convergence rate guarantees. Empirically, experiments on large random matrix games and normal-form subgames of the AI benchmark Diplomacy show that greedy weights outperforms previous methods whenever sampling is used, sometimes by several orders of magnitude.

poster session 5 @ blue 6, poster session 8 @ blue 6, poster session 5, poster session 8

Knowledge transfer from synthetic to real data has been widely studied to mitigate data annotation constraints in various computer vision tasks such as semantic segmentation. However, the study focused on 2D images and its counterpart in 3D point clouds segmentation lags far behind due to the lack of large-scale synthetic datasets and effective transfer methods. We address this issue by collecting SynLiDAR, a synthetic LiDAR dataset that contains large-scale point-wise annotated point cloud with accurate geometric shapes and comprehensive semantic classes. Leveraging graphic tools and professionals, SynLiDAR are collected from multiple virtual environments with rich types of scenes and layouts. In addition, we design PCT, a novel point cloud translator that effectively mitigates the gap between synthetic and real point clouds. Specifically, we decompose the synthetic-to-real gap into an appearance component and a sparsity component and handle them separately which improves the point cloud translation greatly. We conducted extensive experiments over three transfer learning setups including data augmentation, semi-supervised domain adaptation and unsupervised domain adaptation. Extensive experiments show that SynLiDAR provides a high-quality data source for studying 3D transfer and the proposed PCT achieves superior point cloud translation consistently across the three setups.

poster session 3 @ red 2, poster session 8 @ red 2, poster session 3, poster session 8

We revisit the setting of fairly allocating indivisible items when agents have different weights representing their entitlements. First, we propose a parameterized family of relaxations for weighted envy-freeness and the same for weighted proportionality; the parameters indicate whether smaller-weight or larger-weight agents should be given a higher priority. We show that each notion in these families can always be satisfied, but any two cannot necessarily be fulfilled simultaneously. We then introduce an intuitive weighted generalization of maximin share fairness and establish the optimal approximation of it that can be guaranteed. Furthermore, we characterize the implication relations between the various weighted fairness notions introduced in this and prior work, and relate them to the lower and upper quota axioms from apportionment.

poster session 5 @ blue 6, poster session 8 @ blue 6, poster session 5, poster session 8

This paper presents a novel collaborative generative modeling (CGM) framework that incentivizes collaboration among self-interested parties to contribute data to a pool for training a generative model (e.g., GAN), from which synthetic data are drawn and distributed to the parties as rewards commensurate to their contributions. Distributing synthetic data as rewards (instead of trained models or money) offers task- and model-agnostic benefits for downstream learning tasks and is less likely to violate data privacy regulation. To realize the framework, we firstly propose a data valuation function using maximum mean discrepancy (MMD) that values data based on its quantity and quality in terms of its closeness to the true data distribution and provide theoretical results guiding the kernel choice in our MMD-based data valuation function. Then, we formulate the reward scheme as a linear optimization problem that when solved, guarantees certain incentives such as fairness in the CGM framework. We devise a weighted sampling algorithm for generating synthetic data to be distributed to each party as reward such that the value of its data and the synthetic data combined matches its assigned reward value by the reward scheme. We empirically show using simulated and real-world datasets that the parties' synthetic data rewards are commensurate to their contributions.

poster session 4 @ red 3, poster session 8 @ red 3, oral session 4 @ red 3, poster session 4, poster session 8, oral session 4

It is well-known that the search algorithms A* and Iterative Deepening A* (IDA*) can fail to solve state-space tasks optimally due to time and memory limits. The former typically fails in memory-restricted scenarios and the latter in time-restricted scenarios. Therefore, several algorithms were proposed to solve state-space tasks optimally using less memory than A* and less time than IDA*, such as A*+IDA*, a hybrid memory-restricted algorithm that combines A* and IDA*. In this paper, we present a hybrid memory-restricted algorithm that combines Partial Expansion A* (PEA*) and IDA*. This new algorithm has two phases, the same structure as the A*+IDA* algorithm. The first phase of PEA*+IDA* runs PEA* until it reaches a memory limit, and the second phase runs IDA* without duplicate detection on each node of PEA*'s Open. First, we present a model that shows how PEA*+IDA* can perform better than A*+IDA* although pure PEA* usually makes more expansions than pure A*. Later, we perform an experimental evaluation using three memory limits and show that compared to A*+IDA* on classical planning domains, PEA*+IDA* has higher coverage and expands fewer nodes. Finally, we experimentally analyze both algorithms and show that having higher F-limits and better priority-queue composition given by PEA* have a considerable impact on the performance of the algorithms.

poster session 4 @ blue 4, poster session 8 @ blue 4, poster session 4, poster session 8

A significant challenge in credit risk models for underwriting is the presence of bias in model training data. When most credit risk models are built using only applicants who had been funded for credit, such non-random sampling predominantly influenced by credit policymakers and previous loan performances may introduce sampling bias to the models, and thus alter their prediction of default on loan repayment when screening applications from prospective borrowers. In this paper, we propose a novel data augmentation method that aims to identify and pseudo-label parts of the historically declined loan applications to mitigate sampling bias in the training data. We also introduce a new measure to assess the performance from the business perspective, loan application approval rates at various loan default rate levels. Our proposed methods were compared to the original supervised learning model and the traditional sampling issue remedy techniques in the industry. The experiment and early production results from deployed model show that self-training method with calibrated probability as data augmentation selection criteria improved the ability of credit scoring to differentiate default loan applications and, more importantly, can increase loan approval rate up to 8.8\%, while keeping similar default rate comparing to baselines. The results demonstrate practical implications on how future underwriting model development processes should follow.

poster session 4 @ red 6, poster session 8 @ red 6, poster session 4, poster session 8

Web search engines focus on serving highly relevant results within hundreds of milliseconds. Pre-trained language transformer models such as BERT are therefore hard to use in this scenario due to their high computational demands. We present our real-time approach to the document ranking problem leveraging a BERT-based siamese architecture. The model is already deployed in a commercial search engine and it improves production performance by more than 3%. For further research and evaluation, we release DaReCzech, a unique data set of 1.6 million Czech user query-document pairs with manually assigned relevance levels. We also release Small-E-Czech, an Electra-small language model pre-trained on a large Czech corpus. We believe this data will support endeavours both of search relevance and multilingual-focused research communities.

poster session 1 @ red 6, poster session 8 @ red 6, poster session 1, poster session 8

Simplicial neural networks (SNNs) have recently emerged as a new direction in graph learning which expands the idea of convolutional architectures from node space to simplicial complexes on graphs. Instead of predominantly assessing pairwise relations among nodes as in the current practice, simplicial complexes allow us to describe higher-order interactions and multi-node graph structures. By building upon connection between the convolution operation and the new block Hodge-Laplacian, we propose the first SNN for link prediction. Our new Block Simplicial Complex Neural Networks (BScNets) model generalizes existing graph convolutional network (GCN) frameworks by systematically incorporating salient interactions among multiple higher-order graph structures of different dimensions. We discuss theoretical foundations behind BScNets and illustrate its utility for link prediction on eight real-world and synthetic datasets. Our experiments indicate that BScNets outperforms the state-of-the-art models by a significant margin while maintaining low computation costs. Finally, we show utility of BScNets as a new promising alternative for tracking spread of infectious diseases such as COVID-19 and measuring the effectiveness of the healthcare risk mitigation strategies.

poster session 1 @ blue 1, poster session 8 @ blue 1, poster session 1, poster session 8

In this work we propose a scheme, called XDC, that uses adversarial learning to train an adaptive cross domain clustering model. XDC trains a classifier on a labeled dataset and assigns labels to an unlabeled dataset. We benefit from adversarial learning such that the target dataset takes part in the training. We also use an existing image classifiers in a plug-and-play fashion (i.e, it can be replaced with any other image classifier). Unlike existing works we update the parameters of the encoder and expose the target dataset to the model during training. We apply our model on two face dataset and one non-face dataset and obtain comparable results with state-of-the-art face clustering models.

poster session 4 @ red 3, poster session 8 @ red 3, poster session 4, poster session 8

Clustering-based unsupervised domain adaptive (UDA) person re-identification (ReID) reduces exhaustive annotations. However, owing to unsatisfactory feature embedding and imperfect clustering, pseudo labels for target domain data inherently contain an unknown proportion of wrong ones, which would mislead feature learning. In this paper, we propose an approach named probabilistic uncertainty guided progressive label refinery (P2LR) for domain adaptive person re-identification. First, we propose to model the labeling uncertainty with the probabilistic distance along with ideal single-peak distributions. A quantitative criterion is established to measure the uncertainty of pseudo labels and facilitate the network training. Second, we explore a progressive strategy for refining pseudo labels. With the uncertainty-guided alternative optimization, we balance between the exploration of target domain data and the negative effects of noisy labeling. On top of a strong baseline, we obtain significant improvements and achieve the state-of-the-art performance on four UDA ReID benchmarks. Specifically, our method outperforms the baseline by 6.5% mAP on the Duke2Market task, while surpassing the state-of-the-art method by 2.5% mAP on the Market2MSMT task.

poster session 3 @ red 1, poster session 8 @ red 1, poster session 3, poster session 8

A conversation is a sequence of utterances, where each utterance plays a specific discourse role while expressing a particular emotion. This paper proposes a novel method to exploit latent discourse role information of an utterance to determine the emotion it conveys in a conversation. Specifically, we use a variant of the Variational-Autoencoder (VAE) to model the context-aware latent discourse roles of each utterance in an unsupervised way. The latent discourse role representation further equips the utterance representation with a salient clue for more accurate emotion recognition. Our experiments show that our proposed method beats the best-reported performances on three public Emotion Recognition in Conversation datasets. This proves that the discourse role information of an utterance plays an important role in the emotion recognition task, which no previous work has studied.

poster session 3 @ red 5, poster session 8 @ red 5, poster session 3, poster session 8

Deep learning-based approaches have shown remarkable performance in the 3D object detection task. However, they suffer from a catastrophic performance drop on the originally trained classes when incrementally learning new classes without revisiting the old data. This "catastrophic forgetting" phenomenon impedes the deployment of 3D object detection approaches in real-world scenarios, where continuous learning systems are needed. In this paper, we study the unexplored yet important class-incremental 3D object detection problem and present the first solution - SDCoT, a novel static-dynamic co-teaching method. Our SDCoT alleviates the catastrophic forgetting of old classes via a static teacher, which provides pseudo annotations for old classes in the new samples and regularizes the current model by extracting previous knowledge with a distillation loss. At the same time, SDCoT consistently learns the underlying knowledge from new data via a dynamic teacher. We conduct extensive experiments on two benchmark datasets and demonstrate the superior performance of our SDCoT over baseline approaches in several incremental learning scenarios.

poster session 3 @ red 2, poster session 8 @ red 2, oral session 3 @ red 2, poster session 3, poster session 8, oral session 3

Digitization of newspapers is of interest for many reasons including preservation of history, accessibility and search ability, etc. While digitization of documents such as scientific articles and magazines is prevalent in literature, one of the main challenges for digitization of newspaper lies in its complex layout (e.g. articles spanning multiple columns, text interrupted by images) analysis, which is necessary to preserve human read-order. This work provides a major breakthrough in the digitization of newspapers on three fronts: first, releasing a dataset of 3000 fully-annotated, real-world newspaper images from 21 different U.S. states representing an extensive variety of complex layouts for document layout analysis; second, proposing layout segmentation as a precursor to existing optical character recognition (OCR) engines, where multiple state-of-the-art image segmentation models and several post-processing methods are explored for document layout segmentation; third, providing a thorough and structured evaluation protocol for isolated layout segmentation and end-to-end OCR.

poster session 7 @ red 6, poster session 8 @ red 6, poster session 7, poster session 8

In this paper, we propose a simple unsupervised graph representation learning method to conduct effective and efficient contrastive learning. Specifically, the proposed multiplet loss explores the complementary information between the structural information and neighbor information to enlarge the inter-class variation, as well as adds an upper bound loss to achieve the finite distance between positive embeddings and anchor embeddings for reducing the intra-class variation. As a result, both enlarging inter-class variation and reducing intra-class variation result in small generalization error, thereby obtaining an effective model. Furthermore, our method removes widely used data augmentation and discriminator from previous graph contrastive learning methods, meanwhile available to output low-dimensional embeddings, leading to an efficient model. Experimental results on various real-world datasets show the effectiveness and efficiency of our method, compared to state-of-the-art methods.

poster session 3 @ blue 2, poster session 8 @ blue 2, oral session 8 @ blue 2, poster session 3, poster session 8, oral session 8

Hierarchical reinforcement learning has focused on discovering temporally extended actions, such as options, that can provide benefits in problems requiring extensive exploration. One promising approach that learns these options end-to-end is the option-critic (OC) framework. We examine and show in this paper that OC does not decompose a problem into simpler sub-problems, but instead increases the size of the search over policy space with each option considering the entire state space during learning. This issue can result in practical limitations of this method, including sample inefficient learning. To address this problem, we introduce Context-Specific Representation Abstraction for Deep Option Learning (CRADOL), a new framework that considers both temporal abstraction and context-specific representation abstraction to effectively reduce the size of the search over policy space. Specifically, our method learns a factored belief state representation that enables each option to learn a policy over only a subsection of the state space. We test our method against hierarchical, non-hierarchical, and modular recurrent neural network baselines, demonstrating significant sample efficiency improvements in challenging partially observable environments.

poster session 1 @ blue 1, poster session 8 @ blue 1, poster session 1, poster session 8

Robust adversarial reinforcement learning is an effective method to train agents to manage uncertain disturbance and modeling errors in real environments. However, for systems that are sensitive to disturbances or those that are difficult to stabilize, it is easier to learn a powerful adversary than establish a stable control policy. An improper strong adversary can destabilize the system, introduce biases in the sampling process, make the learning process unstable, and even reduce the robustness of the policy. In this study, we consider the problem of ensuring system stability during training in the adversarial reinforcement learning architecture. The dissipative principle of robust $H_\infty$ control is extended to the Markov Decision Process, and robust stability constraints are obtained based on $L_2$ gain performance in the reinforcement learning system. Thus, we propose a dissipation-inequation-constraint-based adversarial reinforcement learning architecture. This architecture ensures the stability of the system during training by imposing constraints on the normal and adversarial agents. Theoretically, this architecture can be applied to a large family of deep reinforcement learning algorithms. Results of experiments in MuJoCo and GymFc environments show that our architecture effectively improves the robustness of the controller against environmental changes and adapts to more powerful adversaries. Results of the flight experiments on a real quadcopter indicate that our method can directly deploy the policy trained in the simulation environment to the real environment, and our controller outperforms the PID controller based on hardware-in-the-loop. Both our theoretical and empirical results provide new and critical outlooks on the adversarial reinforcement learning architecture from a rigorous robust control perspective.

poster session 1 @ blue 1, poster session 8 @ blue 1, poster session 1, poster session 8

Graph Convolutional Networks (GCNs) have been widely applied in various fields due to their significant power on processing graph-structured data. Typical GCN and its variants work under a homophily assumption (i.e., nodes with same class are prone to connect to each other), while ignoring the heterophily which exists in many real-world networks (i.e., nodes with different classes tend to form edges). Existing methods deal with heterophily by mainly aggregating higher-order neighborhoods or combing the immediate representations, which leads to noise and irrelevant information in the result. But these methods did not change the propagation mechanism which works under homophily assumption (that is a fundamental part of GCNs). This makes it difficult to distinguish the representation of nodes from different classes. To address this problem, in this paper we design a novel propagation mechanism, which can automatically change the propagation and aggregation process according to homophily or heterophily between node pairs. To adaptively learn the propagation process, we introduce two measurements of homophily degree between node pairs, which is learned based on topological and attribute information, respectively. Then we incorporate the learnable homophily degree into the graph convolution framework, which is trained in an end-to-end schema, enabling it to go beyond the assumption of homophily. More importantly, we theoretically prove that our model can constrain the similarity of representations between nodes according to their homophily degree. Experiments on seven real-world datasets demonstrate that this new approach outperforms the state-of-the-art methods under heterophily or low homophily, and gains competitive performance under homophily.

poster session 3 @ blue 2, poster session 8 @ blue 2, oral session 3 @ blue 2, poster session 3, poster session 8, oral session 3

We study the problem of fair and efficient allocation of a set of indivisible chores to agents with additive cost functions. We consider the popular fairness notion of envy-freeness up to one good (EF1) with the efficiency notion of Pareto-optimality (PO). While it is known that EF1+PO allocations exists and can be computed in pseudo-polynomial time in the case of goods, the same problem is open for chores. Our first result is a strongly polynomial-time algorithm for computing an EF1+PO allocation for bivalued instances, where agents have (at most) two disutility values for the chores. To the best of our knowledge, this is the first non-trivial class of chores to admit an EF1+PO allocation and an efficient algorithm for its computation. We also study the problem of computing an envy-free (EF) and PO allocation for the case of divisible chores. While the existence of EF+PO allocation is known via competitive equilibrium with equal incomes, its efficient computation is open. Our second result shows that for bivalued instances, an EF+PO allocation can be computed in strongly polynomial-time.

poster session 5 @ blue 6, poster session 8 @ blue 6, poster session 5, poster session 8

Recovering a dense depth image from sparse LiDAR scans is a challenging task. Despite the popularity of color-guided methods for sparse-to-dense depth completion, they treated pixels equally during optimization, ignoring the uneven distribution characteristics in the sparse depth map and the accumulated outliers in the synthesized ground truth. In this work, we introduce uncertainty-driven loss functions to improve the robustness of depth completion and handle the uncertainty in depth completion. Specifically, we propose an explicit uncertainty formulation for robust depth completion with Jeffrey's prior. A parametric uncertain-driven loss is introduced and translated to new loss functions that are robust to noisy or missing data. Meanwhile, we propose a multiscale joint prediction model that can simultaneously predict depth and uncertainty maps. The estimated uncertainty map is also used to perform adaptive prediction on the pixels with high uncertainty, leading to a residual map for refining the completion results. Our method has been tested on KITTI Depth Completion Benchmark and achieved the state-of-the-art robustness performance in terms of MAE, IMAE, and IRMSE metrics.

poster session 3 @ red 2, poster session 8 @ red 2, oral session 3 @ red 2, poster session 3, poster session 8, oral session 3

Multi-Agent Path Finding (MAPF) is the problem of planning collision-free paths for multiple agents in a shared environment. In this paper, we propose a novel algorithm MAPF-LNS2 based on large neighborhood search for solving MAPF efficiently. Starting from a set of paths that contains collisions, MAPF-LNS2 repeatedly selects a subset of colliding agents and replans their paths to reduce the number of collisions until the paths become collision-free. We compare MAPF-LNS2 against a variety of state-of-the-art MAPF algorithms, including Prioritized Planning with random restarts, EECBS, and PPS, and show that MAPF-LNS2 runs significantly faster than them while still providing near-optimal solutions in most cases. With a time limit of just 5 minutes, MAPF-LNS2 managed to solve 80% of the random-scenario instances on all 33 maps with the largest numbers of agents from the MAPF benchmark, which clearly defines the new state-of-the-art suboptimal MAPF algorithm.

poster session 4 @ red 3, poster session 8 @ red 3, poster session 4, poster session 8

Unsupervised Neural Machines Translation is the most ideal method to apply to Cantonese and Chinese translation because parallel data is scarce in this language pair. In this paper, we proposed a method that combined a modified cross-lingual language model and performed layer to layer attention on unsupervised neural machine translation. In our experiments, we observed that our proposed method does improve the Cantonese to Chinese and Chinese to Cantonese translation by 1.088 and 0.394 BLEU scores. We finally developed a web service based on our ideal approach to provide Cantonese to Chinese Translation and vice versa.

poster session 1 @ red 4, poster session 8 @ red 4, poster session 1, poster session 8

Neural network quantization has become increasingly popular due to efficient memory consumption and faster computation resulting from bitwise operations on the quantized networks. Even though they exhibit excellent generalization capabilities, their robustness properties are not well-understood. In this work, we systematically study the robustness of quantized networks against gradient based adversarial attacks and demonstrate that these quantized models suffer from gradient vanishing issues and show a fake sense of robustness. By attributing gradient vanishing to poor forward-backward signal propagation in the trained network, we introduce a simple temperature scaling approach to mitigate this issue while preserving the decision boundary. Despite being a simple modification to existing gradient based adversarial attacks, experiments on multiple image classification datasets with multiple network architectures demonstrate that our temperature scaled attacks obtain near-perfect success rate on quantized networks while outperforming original attacks on adversarially trained models as well as floating-point networks.

poster session 3 @ blue 3, poster session 8 @ blue 3, poster session 3, poster session 8

Description logic ontologies serve to model classifications and structural relationships, and to represent and reason about domain knowledge. Modeling the basic classification of abstract algebraic structures as an ontology demonstrates the difficulties presented by their logical semantics, and shed light on the limitations to accurately model further topics in algebra and related mathematical domains.

poster session 4 @ blue 5, poster session 8 @ blue 5, poster session 4, poster session 8

Modern multi-agent robotic systems increasingly require scalable, robust and persistent Multi-Agent Path Finding (MAPF) with performance guarantees. While many MAPF solvers that provide some of these properties exist, none provides them all. To fill this need, we propose a new MAPF framework, the shard system. A shard system partitions the workspace into geographic regions, called shards, linked by a novel system of buffers. Agents are routed optimally within a shard by a local controller to local goals set by a global controller. The buffer system novelly allows shards to plan with perfect parallelism, providing scalability. A novel global controller algorithm can rapidly generate an inter-shard routing plan for thousands of agents while minimizing the traffic routed through any shard. A novel workspace partitioning algorithm produces shards small enough to replan rapidly. These innovations allow a shard system to adjust its routing plan in real time if an agent is delayed or assigned a new goal, enabling robust, persistent MAPF. A shard system's local optimality and optimized inter-shard routing bring the sum-of-costs of its solutions to single-shot MAPF problems to < 20-60% of optimal on a diversity of workspaces. Its scalability allows it to plan paths for 1000s of agents in seconds. If any of their goals change or move actions fails, a shard system can replan in under a second.

poster session 4 @ red 3, poster session 8 @ red 3, poster session 4, poster session 8

One of the ways to make reinforcement learning (RL) more efficient is by utilizing human advice. Because human advice is expensive, the central question in advice-based reinforcement learning is, how to decide in which states the agent should ask for advice. To approach this challenge, various advice strategies have been proposed. Although all of these strategies distribute advice more efficiently than naive strategies, they rely solely on the agent's estimate of the action-value function, and therefore, are rather inefficient when this estimate is not accurate, in particular, in the early stages of the learning process. To address this weakness, we present an approach to advice-based RL, in which the human’s role is not limited to giving advice in chosen states, but also includes hinting a-priori, before the learning procedure, in which sub-domains of the state space the agent might require more advice. For this purpose we use the concept of critical: states in which choosing the proper action is more important than in other states.

poster session 4 @ blue 5, poster session 8 @ blue 5, poster session 4, poster session 8

Recent work has shown that Binarized Neural Networks (BNNs) are able to greatly reduce computational costs and memory footprints, facilitating model deployment on resource-constrained devices. However, in comparison to their full-precision counterparts, BNNs suffer from severe accuracy degradation. Research aiming to reduce this accuracy gap has thus far largely focused on specific network architectures with few or no 1 × 1 convolutional layers, for which standard binarization methods do not work well. Because 1 × 1 convolutions are common in the design of modern architectures (e.g. GoogleNet, ResNet, DenseNet), it is crucial to develop a method to binarize them effectively for BNNs to be more widely adopted. In this work, we propose an “Elastic-Link” (EL) module to enrich information flow within a BNN by adaptively adding real-valued input features to the subsequent convolutional output features. The proposed EL module is easily implemented and can be used in conjunction with other methods for BNNs. We demonstrate that adding EL to BNNs produces a significant improvement on the challenging large-scale ImageNet dataset. For example, we raise the top-1 accuracy of binarized ResNet26 from 57.9% to 64.0%. EL also aids con-vergence in the training of binarized MobileNet, for which a top-1 accuracy of 56.4% is achieved. Finally, with the integration of ReActNet, it yields a new state-of-the-art result of 71.9% top-1 accuracy.

poster session 4 @ red 3, poster session 8 @ red 3, poster session 4, poster session 8

Local image feature matching under large appearance, viewpoint, and distance changes is challenging yet important. Conventional methods detect and match tentative local features across the whole images, with heuristic consistency checks to guarantee reliable matches. In this paper, we introduce a novel Overlap Estimation method conditioned on image pairs with TRansformer, named OETR, to constrain local feature matching in the commonly visible region. OETR performs overlap estimation in a two-step process of feature correlation and then overlap regression. As a preprocessing module, OETR can be plugged into any existing local feature detection and matching pipeline, to mitigate potential view angle or scale variance. Intensive experiments show that OETR can boost state-of-the-art local feature matching performance substantially, especially for image pairs with small shared regions. Code will be available upon publication.

poster session 3 @ red 2, poster session 8 @ red 2, poster session 3, poster session 8

Deep unsupervised hashing has been appreciated in the regime of image retrieval. However, most prior arts failed to detect the semantic components and their relationships behind the images, which makes them lack discriminative power. To make up the defect, we propose a novel \textbf{Deep Semantic Components Hashing (DSCH)}, which involves a common sense that an image normally contains a bunch of semantic components with homology and co-occurrence relationships. Based on this prior, DSCH regards the semantic components as latent variables under the Expectation-Maximization framework and designs a two-step iterative algorithm with the objective of maximum likelihood of training data. Firstly, DSCH constructs a semantic component structure by uncovering the fine-grained semantics components of images with a Gaussian Mixture Modal~(GMM), where an image is represented as a mixture of multiple components, and the semantics co-occurrence are exploited. Besides, coarse-grained semantics components, are discovered by considering the homology relationships between fine-grained components, and the hierarchy organization is then constructed. Secondly, DSCH makes the images close to their semantic component centers at both fine-grained and coarse-grained levels, and also makes the images share similar semantic components close to each other. Extensive experiments on three benchmark datasets demonstrate that the proposed hierarchical semantic components indeed facilitate the hashing model to achieve superior performance.

poster session 3 @ red 1, poster session 8 @ red 1, oral session 3 @ red 1, poster session 3, poster session 8, oral session 3

We consider language modelling (LM) as a multi-label structured prediction task by re-framing training from solely predicting a single ground-truth word to ranking a set of words which could continue a given context. To avoid annotating top-k ranks, we generate them using pre-trained LMs: GPT-2, BERT, and Born-Again models. This leads to a rank-based form of knowledge distillation (KD). We also develop a method using $N$-grams to create a non-probabilistic teacher which generates the ranks without the need of a pre-trained LM. We confirm the hypotheses: that we can treat LMing as a ranking task and that we can do so without the use of a pre-trained LM. We show that rank-based KD generally gives a modest improvement to perplexity (PPL) -- though often with statistical significance -- when compared to Kullback–Leibler-based KD. Surprisingly, given the naivety of the method, the $N$-grams act as competitive teachers and achieve similar performance as using either BERT or a Born-Again model teachers. Unsurprisingly, GPT-2 always acts as the best teacher. Using it and a Transformer-XL student on Wiki-02, rank-based KD reduces a cross-entropy baseline from 65.27 to 55.94 and against a KL-based KD of 56.70.

poster session 1 @ red 4, poster session 8 @ red 4, oral session 8 @ red 4, poster session 1, poster session 8, oral session 8

A key trait of daily conversations between individuals is the ability to express empathy towards others, and exploring ways to implement empathy is a crucial step towards human-like dialogue systems. Previous approaches on this topic mainly focus on detecting and utilizing the user's emotion for generating empathetic responses. However, since empathy includes both aspects of affection and cognition, we argue that in addition to identifying the user's emotion, cognitive understanding of the user's situation should also be considered. To this end, we propose a novel approach for empathetic response generation, which leverages commonsense to draw more information about the user's situation and uses this additional information to further enhance the empathy expression in generated responses. We evaluate our approach on EmpatheticDialogues, which is a widely-used benchmark dataset for empathetic response generation. Empirical results demonstrate that our approach outperforms the baseline models in both automatic and human evaluations and can generate more informative and empathetic responses.

poster session 3 @ red 5, poster session 8 @ red 5, oral session 3 @ red 5, poster session 3, poster session 8, oral session 3

Modeling the preferences of agents over a set of alternatives is a principal concern in many areas. The dominant approach has been to find a single reward/utility function with the property that alternatives yielding higher rewards are preferred over alternatives yielding lower rewards. However, in many settings, preferences are based on multiple—often competing—objectives; a single reward function is not adequate to represent such preferences. This paper proposes a method for inferring multi-objective reward-based representations of an agent's observed preferences. We model the agent's priorities over different objectives as entering lexicographically, so that objectives with lower priorities matter only when the agent is indifferent with respect to objectives with higher priorities. We offer two example applications in healthcare—one inspired by cancer treatment, the other inspired by organ transplantation—to illustrate how the lexicographically-ordered rewards we learn can provide a better understanding of a decision-maker's preferences and help improve policies when used in reinforcement learning.

poster session 4 @ blue 5, poster session 8 @ blue 5, poster session 4, poster session 8

Simultaneous Localization and Mapping (SLAM) and Autonomous Driving are becoming increasingly more important in recent years. Point cloud-based large scale place recognition is the spine of them. While many models have been proposed and have achieved acceptable performance by learning short-range local features, they always skip long-range contextual properties. Moreover, the model size also becomes a serious shackle for their wide applications. To overcome these challenges, we propose a super light-weight network model termed SVT-Net. On top of the highly efficient 3D Sparse Convolution (SP-Conv), an Atom-based Sparse Voxel Transformer (ASVT) and a Cluster-based Sparse Voxel Transformer (CSVT) are proposed respectively to learn both short-range local features and long-range contextual features. Consisting of ASVT and CSVT, SVT-Net can achieve state-of-the-art performance in terms of both recognition accuracy and running speed with a super-light model size (0.9M parameters). Meanwhile, for the purpose of further boosting efficiency, we introduce two simplified versions, which also achieve state-of-the-art performance and further reduce the model size to 0.8M and 0.4M respectively.

poster session 3 @ red 2, poster session 8 @ red 2, poster session 3, poster session 8

Generating absent keyphrases, which do not appear in the input document, is challenging in the keyphrase prediction task. Most previous works treat the problem as an autoregressive sequence-to-sequence generation task, which demonstrates promising results for generating grammatically correct and fluent absent keyphrases. However, such an end-to-end process with a complete data-driven manner is unconstrained, which is prone to generate keyphrases inconsistent with the input document. In addition, the existing autoregressive decoding method makes the generation of keyphrases must be done from left to right, leading to slow speed during inference. In this paper, we propose a constrained absent keyphrase generation method in a prompt-based learning fashion. Specifically, the prompt will be created firstly based on the keywords, which are defined as the overlapping words between absent keyphrase and document. Then, a mask-predict decoder is used to complete the absent keyphrase on the constraint of prompt. Experiments on keyphrase generation benchmarks have demonstrated the effectiveness of our approach. In addition, we evaluate the performance of constrained absent keyphrases generation from an information retrieval perspective. The result shows that our approach can generate more consistent keyphrases, which can improve document retrieval performance. What’s more, with a non-autoregressive decoding manner, our model can speed up the absent keyphrase generation by 8.67× compared with the autoregressive method.

poster session 3 @ red 5, poster session 8 @ red 5, poster session 3, poster session 8

Assessing the performance of Generative Adversarial Networks (GANs) has been an important topic due to its practical significance. Although several evaluation metrics have been proposed, they generally assess the quality of the whole generated image distribution. For Reference-guided Image Synthesis (RIS) tasks, i.e., rendering a source image in the style of another reference image, where assessing the quality of a single generated image is crucial, these metrics are not applicable. In this paper, we propose a general learning-based framework, Reference-guided Image Synthesis Assessment (RISA) to quantitatively evaluate the quality of a single generated image. Notably, the training of RISA does not require human annotations. In specific, the training data for RISA are acquired by the intermediate models from the training procedure in RIS, and weakly annotated by the number of models' iterations, based on the positive correlation between image quality and iterations. As this annotation is too coarse as a supervision signal, we introduce two techniques: 1) a pixel-wise interpolation scheme to refine the coarse labels, and 2) multiple binary classifiers to replace a naïve regressor. In addition, an unsupervised contrastive loss is introduced to effectively capture the style similarity between a generated image and its reference image. Empirical results on various datasets demonstrate that RISA is highly consistent with human preference and transfers well across models.

poster session 3 @ red 1, poster session 8 @ red 1, oral session 8 @ red 1, poster session 3, poster session 8, oral session 8

Federated learning (FL) enables distribution of machine learning workloads from the cloud to resource-limited edge devices. Unfortunately, current deep networks remain not only too compute-heavy for inference and training on edge devices, but also too large for communicating updates over bandwidth-constrained networks. In this paper, we develop, implement, and experimentally validate a novel FL framework termed Federated Dynamic Sparse Training (FedDST) by which complex neural networks can be deployed and trained with substantially improved efficiency in both on-device computation and in-network communication. At the core of FedDST is a dynamic process that extracts and trains sparse sub-networks from the target full network. With this scheme, "two birds are killed with one stone:'' instead of full models, each client performs efficient training of its own sparse networks, and only sparse networks are transmitted between devices and the cloud. Furthermore, our results reveal that the dynamic sparsity during FL training more flexibly accommodates local heterogeneity in FL agents than the fixed, shared sparse masks. Moreover, dynamic sparsity naturally introduces an "in-time self-ensembling effect'' into the training dynamics, and improves the FL performance even over dense training. In a realistic and challenging non i.i.d. FL setting, FedDST consistently outperforms competing algorithms in our experiments: for instance, at any fixed upload data cap on non-iid CIFAR-10, it gains an impressive accuracy advantage of 10% over FedAvgM when given the same upload data cap; the accuracy gap remains 3% even when FedAvgM is given 2 times the upload data cap, further demonstrating efficacy of FedDST. Code is available at: https://github.com/bibikar/feddst.

poster session 3 @ blue 3, poster session 8 @ blue 3, poster session 3, poster session 8

In Mixed Integer Linear Programming (MIP), a (strong) \textit{backdoor} is a ``small" subset of an instance's integer variables with the following property: in a branch-and-bound procedure, the instance can be solved to global optimality by branching \textit{only} on the variables in the backdoor. Constructing datasets of pre-computed backdoors for widely used MIP benchmark sets or particular problem families can enable new questions around novel structural properties of a MIP, or explain why a problem that is hard in theory can be solved efficiently in practice. Existing algorithms for finding backdoors rely on sampling candidate variable subsets in various ways, an approach which has demonstrated the existence of backdoors for some instances from MIPLIB2003 and MIPLIB2010. However, these algorithms fall short of consistently succeeding at the task due to an imbalance between exploration and exploitation. We propose \bamcts, a Monte Carlo Tree Search framework for finding backdoors to MIPs. Extensive algorithmic engineering, hybridization with traditional MIP concepts, and close integration with the CPLEX solver have enabled our method to outperform baselines on MIPLIB2017 instances, finding backdoors more frequently and more efficiently.

poster session 4 @ blue 4, poster session 8 @ blue 4, oral session 8 @ blue 4, poster session 4, poster session 8, oral session 8

In order to fully perceive the surrounding environment, many intelligent robots and self-driving cars are equipped with a multi-camera system. Based on this system, the structure-from-motion (SfM) technology is used to realize scene reconstruction, but the fixed relative poses between cameras in the multi-camera system are usually not considered. This paper presents a tailor-made multi-camera based motion averaging system, where the fixed relative poses are utilized to improve the accuracy and robustness of SfM. Our approach starts by dividing the images into reference images and non-reference images, and edges in view-graph are divided into four categories accordingly. Then, a multi-camera based rotating averaging problem is formulated and solved in two stages, where an iterative re-weighted least squares scheme is used to deal with outliers. Finally, a multi-camera based translation averaging problem is formulated and a l1-norm based optimization scheme is proposed to compute the relative translations of multi-camera system and reference camera positions simultaneously. Experiments demonstrate that our algorithm achieves superior accuracy and robustness on various data sets compared to the state-of-the-art methods.

poster session 3 @ red 2, poster session 8 @ red 2, oral session 8 @ red 2, poster session 3, poster session 8, oral session 8

Meta reinforcement learning (meta-RL) aims to learn a policy solving a set of training tasks simultaneously and quickly adapting to new tasks. It requires massive amounts of data drawn from training tasks to infer the common structure shared among tasks. Without heavy reward engineering, the sparse rewards in long-horizon tasks exacerbate the problem of sample efficiency in meta-RL. Another challenge in meta-RL is the discrepancy of difficulty level among tasks, which might cause one easy task dominating learning of the shared policy and thus preclude policy adaptation to new tasks. This work introduces a novel objective function to learn an action translator among training tasks. We theoretically verify that the value of the transferred policy with the action translator can be close to the value of the source policy and our objective function (approximately) upper bounds the value difference. We propose to combine the action translator with context-based meta-RL algorithms for better data collection and moreefficient exploration during meta-training. Our approach em-pirically improves the sample efficiency and performance ofmeta-RL algorithms on sparse-reward tasks.

poster session 1 @ blue 1, poster session 8 @ blue 1, poster session 1, poster session 8

Online reinforcement learning (RL) algorithms are often difficult to deploy in complex human-facing applications as they may learn slowly and have poor early performance. To address this, we introduce a practical algorithm for incorporating human insight to speed learning. Our algorithm, Constraint Sampling Reinforcement Learning (CSRL), incorporates prior domain knowledge as constraints/restrictions on the RL policy. It takes in multiple potential policy constraints to maintain robustness to misspecification of individual constraints while leveraging helpful ones to learn quickly. Given a base RL learning algorithm (ex. UCRL, DQN, Rainbow) we propose an upper confidence with elimination scheme that leverages the relationship between the constraints, and their observed performance, to adaptively switch among them. We instantiate our algorithm with DQN-type algorithms and UCRL as base algorithms, and evaluate our algorithm in four environments, including three simulators based on real data: recommendations, educational activity sequencing, and HIV treatment sequencing. In all cases, CSRL learns a good policy faster than baselines.

poster session 1 @ blue 1, poster session 8 @ blue 1, poster session 1, poster session 8

The goal of building intelligent dialogue systems has largely been separately pursued under two paradigms: task-oriented dialogue (TOD) systems, which perform task-specific functions, and open-domain dialogue (ODD) systems, which focus on non-goal-oriented chitchat. The two dialogue modes can potentially be intertwined together seamlessly in the same conversation, as easily done by a friendly human assistant. Such ability is desirable in conversational agents, as the integration makes them more accessible and useful. Our paper addresses this problem of fusing TODs and ODDs in multi-turn dialogues. Based on the popular TOD dataset MultiWOZ, we build a new dataset FusedChat, by rewriting the existing TOD turns and adding new ODD turns. This procedure constructs conversation sessions containing exchanges from both dialogue modes. It features inter-mode contextual dependency, i.e., the dialogue turns from the two modes depend on each other. Rich dependency patterns such as co-reference and ellipsis are included. The new dataset, with 60k new human-written ODD turns and 5k re-written TOD turns, offers a benchmark to test a dialogue model's ability to perform inter-mode conversations. This is a more challenging task since the model has to determine the appropriate dialogue mode and generate the response based on the inter-mode context. However, such models would better mimic human-level conversation capabilities. We evaluate two baseline models on this task, including the classification-based two-stage models and the two-in-one fused models. We publicly release FusedChat (https://github.com/tomyoung903/FusedChat) and the baselines to propel future work on inter-mode dialogue systems.

poster session 3 @ red 5, poster session 8 @ red 5, oral session 8 @ red 5, poster session 3, poster session 8, oral session 8

The goal of a summary is to concisely state the most important information in a document. With this principle in mind, we introduce new reference-free summary evaluation metrics that use a pretrained language model to estimate the information content shared between a document and its summary. These metrics are a modern take on the Shannon Game, a method for summary quality scoring proposed decades ago, where we replace human annotators with language models. We also view these metrics as an extension of BLANC, a recently proposed approach to summary quality measurement based on the performance of a language model with and without the help of a summary. Using transformer based language models, we empirically verify that our metrics achieve state-of-the-art correlation with human judgement of the summary quality dimensions of both coherence and relevance, as well as competitive correlation with human judgement of consistency and fluency.

poster session 1 @ red 4, poster session 8 @ red 4, oral session 1 @ red 4, poster session 1, poster session 8, oral session 1

In dialogue systems, utterances with similar semantics may have distinctive emotions under different contexts. Therefore, modeling long-range contextual emotional relationships with speaker dependency plays a crucial part in dialogue emotion recognition. Meanwhile, distinguishing the different emotion categories is non-trivial since they usually have semantically similar sentiments. To this end, we adopt supervised contrastive learning to make different emotions mutually exclusive to identify similar emotions better. Meanwhile, we utilize an auxiliary response generation task to enhance the model's ability of handling context information, thereby forcing the model to recognize emotions with similar semantics in diverse contexts. To achieve these objectives, we use the pre-trained encoder-decoder model BART as our backbone model since it is very suitable for both understanding and generation tasks. The experiments on four datasets demonstrate that our proposed model obtains significantly more favorable results than the state-of-the-art model in dialogue emotion recognition. The ablation study further demonstrates the effectiveness of supervised contrastive loss and generative loss.

poster session 3 @ red 5, poster session 8 @ red 5, poster session 3, poster session 8

Recent research on graph neural network (GNN) models successfully applied GNNs to classical graph algorithms and combinatorial optimisation problems. This has numerous benefits, such as allowing applications of algorithms when preconditions are not satisfied, or reusing learned models when sufficient training data is not available or can't be generated. Unfortunately, a key hindrance of these approaches is their lack of explainability, since GNNs are black-box models that cannot be interpreted directly. In this work, we address this limitation by applying existing work on concept-based explanations to GNN models. We introduce \textit{concept-bottleneck GNNs}, which rely on a modification to the GNN readout mechanism. Using three case studies we demonstrate that: (i) our proposed model is capable of accurately learning concepts and extracting propositional formulas based on the learned concepts for each target class; (ii) our concept-based GNN models achieve comparative performance with state-of-the-art models; (iii) we can derive global graph concepts, without explicitly providing any supervision on graph-level concepts.

poster session 3 @ blue 2, poster session 8 @ blue 2, poster session 3, poster session 8

While there has been an increasing focus on the use of game theoretic models for autonomous driving, empirical evidence shows that there are still open questions around dealing with the challenges of common knowledge assumptions as well as modeling bounded rationality. To address some of these practical challenges, we develop a framework of generalized dynamic cognitive hierarchy for both modelling naturalistic human driving behavior as well as behavior planning for autonomous vehicles (AV). This framework is built upon a rich model of level-0 behavior through the use of automata strategies, an interpretable notion of bounded rationality through safety and maneuver satisficing, and a robust response for planning. Based on evaluation on two large naturalistic datasets as well as simulation of critical traffic scenarios, we show that i) automata strategies are well suited for level-0 behavior in a dynamic level-k framework, and ii) the proposed robust response to a heterogeneous population of strategic and non-strategic reasoners can be an effective approach for game theoretic planning in AV.

poster session 5 @ blue 6, poster session 8 @ blue 6, poster session 5, poster session 8

In attempts to "explain" predictions of machine learning models, researchers have proposed hundreds of techniques for attributing predictions to features that are deemed important. While these attributions are often claimed to hold the potential to improve human "understanding" of the models, surprisingly little work explicitly evaluates progress towards this aspiration. In this paper, we conduct a crowdsourcing study, where participants interact with deception detection models that have been trained to distinguish between genuine and fake hotel reviews. They are challenged both to simulate the model on fresh reviews, and to edit reviews with the goal of lowering the probability of the originally predicted class. Successful manipulations would lead to an adversarial example. During the training (but not the test) phase, input spans are highlighted to communicate salience. Through our evaluation, we observe that for a linear bag-of-words model, participants with access to the feature coefficients during training are able to cause a larger reduction in model confidence in the testing phase when compared to the no-explanation control. For the BERT-based classifier, popular local explanations do not improve their ability to reduce the model confidence over the no-explanation case. Remarkably, when the explanation for the BERT model is given by the (global) attributions of a linear model trained to imitate the BERT model, people can effectively manipulate the model.

poster session 1 @ red 4, poster session 8 @ red 4, oral session 1 @ red 4, poster session 1, poster session 8, oral session 1

Optimising the approximation of Average Precision (AP) has been widely studied for image retrieval. Limited by the definition of AP, such methods consider both negative and positive instances ranking before each positive instance. However, we claim that only penalizing negative instances before positive ones is enough, because the loss only comes from these negative instances. To this end, we propose a novel loss, namely Penalizing Negative instances before Positive ones (PNP), which can directly minimize the number of negative instances before each positive one. In addition, AP-based methods adopt a fixed and sub-optimal gradient assignment strategy. Therefore, we systematically investigate different gradient assignment solutions via constructing derivative functions of the loss, resulting in PNP-I with increasing derivative functions and PNP-D with decreasing ones. PNP-I focuses more on the hard positive instances by assigning larger gradients to them and tries to make all relevant instances closer. In contrast, PNP-D pays less attention to such instances and slowly corrects them. For most real-world data, one class usually contains several local clusters. PNP-I blindly gathers these clusters while PNP-D keeps them as they were. Therefore, PNP-D is more superior. Experiments on three standard retrieval datasets show consistent results with the above analysis. Extensive evaluations demonstrate that PNP-D achieves the state-of-the-art performance.

poster session 3 @ red 1, poster session 8 @ red 1, poster session 3, poster session 8

Tensor optimization is crucial to massive machine learning and signal processing tasks. In this paper, we consider tensor optimization with a convex and well-conditioned objective function and reformulate it into a nonconvex optimization using the Burer-Monteiro type parameterization. We analyze the local convergence of applying vanilla gradient descent to the factored formulation and establish a local regularity condition under mild assumptions. We also provide a linear convergence analysis of the gradient descent algorithm started in a neighborhood of the true tensor factors. Complementary to the local analysis, this work also characterizes the global geometry of the best rank-one tensor approximation problem and demonstrates that for orthogonally decomposable tensors the problem has no spurious local minima and all saddle points are strict except for the one at zero which is a third-order saddle point.

poster session 4 @ blue 4, poster session 8 @ blue 4, poster session 4, poster session 8

The existence of EFX allocations of goods is a major open problem in fair division, even for additive valuations. The current state of the art is that no setting where EFX allocations are impossible is known, and yet, existence results are known only for very restricted settings, such as: (i) agents with identical valuations, (ii) 2 agents, and (iii) 3 agents with additive valuations. It is also known that EFX exists if one can leave n-1 items unallocated, where n is the number of agents. We develop new techniques that allow us to push the boundaries of the enigmatic EFX problem beyond these known results, and (arguably) to simplify proofs of earlier results. Our main result is that every setting with 4 additive agents admits an EFX allocation that leaves at most a single item unallocated. Beyond our main result, we introduce a new class of valuations, termed nice cancelable, which includes additive, unit-demand, budget-additive and multiplicative valuations, among others. Using our new techniques, we show that both our results and previous results for additive valuations extend to nice cancelable valuations.

poster session 5 @ blue 6, poster session 8 @ blue 6, oral session 8 @ blue 6, poster session 5, poster session 8, oral session 8

Batch Normalization (BN) as an important component assists Deep Neural Networks achieving promising performance for extensive learning tasks by scaling distribution of feature representations within mini-batches. However, the application of BN suffers from performance degradation under the scenario of Unsupervised Domain Adaptation (UDA), since the estimated statistics fail to concurrently describe two different domains. In this paper, we develop a novel normalization technique, named Collaborative Normalization (CoN), for eliminating domain discrepancy and accelerating the model training of neural networks for UDA. Unlike typical strategies only exploiting domain-specific statistics during normalization, our CoN excavates cross-domain knowledge and simultaneously scales features from various domains by mimicking the merits of collaborative representation. Our CoN can be easily plugged into popular neural network backbones for cross-domain learning. On the one hand, \textbf{theoretical analysis} guarantees that models with CoN promote discriminability of feature representations and accelerate convergence rate; on the other hand, \textbf{empirical study} verifies that replacing BN with CoN in popular network backbones effectively improves classification accuracy in most learning tasks across three cross-domain visual benchmarks.

poster session 3 @ blue 3, poster session 8 @ blue 3, poster session 3, poster session 8

Learning to understand grounded language, which connects natural language to percepts, is a critical research area. Prior work in grounded language acquisition has focused primarily on textual inputs. In this work we demonstrate the feasibility of performing grounded language acquisition on paired visual percepts and raw speech inputs. This will allow human-robot interactions in which language about novel tasks and environments is learned from end users, reducing dependence on textual inputs and potentially mitigating the effects of demographic bias found in widely available speech recognition systems. We leverage recent work in self-supervised speech representation models and show that learned representations of speech can make language grounding systems more inclusive towards specific groups while maintaining or even increasing general performance.

poster session 1 @ red 4, poster session 8 @ red 4, poster session 1, poster session 8

Identifying breakdowns in ongoing dialogues helps to improve communication effectiveness. Most prior work on this topic relies on human annotated data and data augmentation to learn a classification model. While quality labeled dialogue data requires human annotation and is usually expensive to obtain, unlabeled data is easier to collect from various sources. In this paper, we propose a novel semi-supervised teacher-student learning framework to tackle this task. We introduce two teachers which are trained on labeled data and perturbed labeled data respectively. We leverage unlabeled data to improve classification in student training where we employ two teachers to refine the labeling of unlabeled data through teacher-student learning in a bootstrapping manner. Through our proposed training approach, the student can achieve improvements over single-teacher performance. Experimental results on the Dialogue Breakdown Detection Challenge dataset DBDC5 and Learning to Identify Follow-Up Questions dataset LIF show that our approach outperforms all previous published approaches as well as other supervised and semi-supervised baseline methods.

poster session 3 @ red 5, poster session 8 @ red 5, oral session 3 @ red 5, poster session 3, poster session 8, oral session 3

Single image generation (SIG), described as generating diverse samples that have the same visual content as the given natural image, is first introduced by SinGAN, which builds a pyramid of GANs to progressively learn the internal patch distribution of the single image. It shows excellent performance in a wide range of image manipulation tasks. However, SinGAN has some limitations. Firstly, due to lack of semantic information, SinGAN cannot handle the object images well as it does on the scene and texture images. Secondly, the independent progressive training scheme is time-consuming and easy to cause artifacts accumulation. To tackle these problems, in this paper, we dig into the single image generation problem and improve SinGAN by fully-utilization of internal and external priors. The main contributions of this paper include: 1) We interpret single image generation from the perspective of the general generative task, that is, to learn a diverse distribution from the Dirac distribution composed of a single image. In order to solve this non-trivial problem, we construct a regularized latent variable model to formulate SIG. To the best of our knowledge, it is the first time to give a clear formulation and optimization goal of SIG, and all the existing methods for SIG can be regarded as special cases of this model. 2) We design a novel Prior-based end-to-end training GAN (PetsGAN), which is infused with internal prior and external prior to overcome the problems of SinGAN. For one thing, we employ the pre-trained GAN model to inject external prior for image generation, which can alleviate the problem of lack of semantic information and generate natural, reasonable and diverse samples, even for the object image. For another, we fully-utilize the internal prior by a differential Patch Matching module and an effective reconstruction network to generate consistent and realistic texture. 3) We construct abundant of qualitative and quantitative experiments on three datasets. The experimental results show our method surpasses other methods on both generated image quality, diversity, and training speed. Moreover, we apply our method to other image manipulation tasks (e.g., style transfer, harmonization) and the results further prove the effectiveness and efficiency of our method.

poster session 3 @ red 1, poster session 8 @ red 1, oral session 8 @ red 1, poster session 3, poster session 8, oral session 8

Recent advances in large-scale pre-training such as GPT-3 allow seemingly high quality text to be generated from a given prompt. However, such generation systems often suffer from problems of hallucinated facts, and are not inherently designed to incorporate useful external information. Grounded generation models appear to offer remedies, but their training typically relies on rarely-available parallel data where corresponding information-relevant documents are provided for context. We propose a framework that alleviates this data constraint by jointly training a grounded generator and document retriever on the language model signal. The model learns to reward retrieval of the documents with the highest utility in generation, and attentively combines them using a Mixture-of-Experts (MoE) ensemble to generate follow-on text. We demonstrate that both generator and retriever can take advantage of this joint training and work synergistically to produce more informative and relevant text in both prose and dialogue generation.

poster session 3 @ red 5, poster session 8 @ red 5, poster session 3, poster session 8

Recent research on model interpretability in natural language processing extensively uses feature scoring methods for identifying which parts of the input are the most important for a model to make a prediction (i.e. explanation or rationale). However, previous research has shown that there is no clear best scoring method across various text classification tasks while practitioners typically have to make several other ad-hoc choices regarding the length and the type of the rationale (e.g. short or long, contiguous or not). Inspired by this, we propose a simple yet effective and flexible method that allows selecting optimally for each data instance: (1) a feature scoring method; (2) the length; and (3) the type of the rationale. Our method is inspired by input erasure approaches to interpretability which assume that the most faithful rationale for a prediction should be the one with the highest difference between the model's output distribution using the full text and the text after removing the rationale as input respectively. Evaluation on four standard text classification datasets shows that our proposed method provides more faithful, comprehensive and highly sufficient explanations compared to using a fixed feature scoring method, rationale length and type. More importantly, we demonstrate that a practitioner is not required to make any ad-hoc choices in order to extract faithful rationales using our approach.

poster session 1 @ red 4, poster session 8 @ red 4, poster session 1, poster session 8

Mean Field Games (MFGs) can potentially scale multi-agent systems to extremely large populations of agents. Yet, most of the literature assumes a single initial distribution for the agents, which limits the practical applications of MFGs. Ma- chine Learning has the potential to solve a wider diversity of MFG problems thanks to generalizations capacities. We study how to leverage these generalization properties to learn policies enabling a typical agent to behave optimally against any population distribution. In reference to the Master equation in MFGs, we coin the term “Master policies” to describe them and we prove that a single Master policy provides a Nash equilibrium, whatever the initial distribution. We propose a method to learn such Master policies. Our approach relies on three ingredients: adding the current population distribution as part of the observation, approximating Master policies with neural networks, and training via Reinforcement Learning and Fictitious Play. We illustrate on numerical examples not only the efficiency of the learned Master policy but also its generalization capabilities beyond the distributions used for training.

poster session 4 @ red 3, poster session 8 @ red 3, poster session 4, poster session 8

Models in the supervised learning framework may capture rich and complex representations over the features that are hard for humans to interpret. Existing methods to explain such models are often specific to architectures and data where the features do not have a time-varying component. In this work, we propose TIME, a method to explain models that are inherently temporal in nature. Our approach (i) uses a model-agnostic permutation-based approach to analyze global feature importance, (ii) identifies the importance of salient features with respect to their temporal ordering as well as localized windows of influence, and (iii) uses hypothesis testing to provide statistical rigor.

poster session 4 @ blue 5, poster session 8 @ blue 5, poster session 4, poster session 8

Point cloud registration is a fundamental step for many tasks. In this paper, we propose a neural network named DetarNet to decouple the translation $t$ and rotation $R$, so as to overcome the performance degradation due to their mutual interference in point cloud registration. First, a Siamese Network based Progressive and Coherent Feature Drift (PCFD) module is proposed to align the source and target points in high-dimensional feature space, and accurately recover translation from the alignment process. Then we propose a Consensus Encoding Unit (CEU) to construct more distinguishable features for a set of putative correspondences. After that, a Spatial and Channel Attention (SCA) block is adopted to build a classification network for finding good correspondences. Finally, the rotation is obtained by Singular Value Decomposition (SVD). In this way, the proposed network decouples the estimation of translation and rotation, resulting in better performance for both of them. Experimental results demonstrate that the proposed DetarNet improves registration performance on both indoor and outdoor scenes. Our code will be available in \url{https://github.com/ZhiChen902/DetarNet}.

poster session 3 @ red 2, poster session 8 @ red 2, oral session 3 @ red 2, poster session 3, poster session 8, oral session 3

In bounded-suboptimal heuristic search, one attempts to find a solution that costs no more than a prespecified factor of optimal as quickly as possible. This is an important setting, as it admits faster-than-optimal solving while retaining some control over solution cost. In this paper, we investigate several new algorithms for bounded-suboptimal search, including novel variants of EES and DPS, the two most prominent previous proposals, and methods inspired by recent work in bounded-cost search that leverages uncertainty estimates of the heuristic. We perform what is, to our knowledge, the most comprehensive empirical comparison of bounded-suboptimal search algorithms to date, including both search and planning benchmarks, and we find that one of the new algorithms, a simple alternating queue scheme, significantly outperforms previous work.

poster session 4 @ blue 4, poster session 8 @ blue 4, oral session 4 @ blue 4, poster session 4, poster session 8, oral session 4

Heterogeneity across clients in federated learning (FL) usually hinders the optimization convergence and generalization performance when the aggregation of clients' knowledge occurs in the gradient space. For example, clients may differ in terms of data distribution, network latency, input/output space, and/or model architecture, which can easily lead to the misalignment of their local gradients. To improve the tolerance to heterogeneity, we propose a novel federated prototype learning (FedProto) framework in which the clients and server communicate the abstract class prototypes instead of the gradients. FedProto aggregates the local prototypes collected from different clients, and then sends the global prototypes back to all clients to regularize the training of local models. The training on each client aims to minimize the classification error on the local data while keeping the resulting local prototypes sufficiently close to the corresponding global ones. Moreover, we provide a theoretical analysis to the convergence rate of FedProto under non-convex objectives. In experiments, we propose a benchmark setting tailored for heterogeneous FL, with FedProto outperforming several recent FL approaches on multiple datasets.

poster session 3 @ blue 3, poster session 8 @ blue 3, poster session 3, poster session 8

Convolutional neural network (CNN) based image segmentation has been widely used in analyzing medical images and benefited many real-world disease diagnosis applications. However, existing advanced CNN-based medical image segmentation models usually contain numerous parameters that require massive computation and memory, limiting the applicability of these models in the data-constrained or hardware-constrained environments. By leveraging the recently proposed neural architecture search (NAS), this paper presents a novel approach, dubbed Thrifty NAS, to design computation and memory-efficient models for medical image segmentation automatically. The searched models by Thrifty NAS are with much fewer parameters while retaining competitive performance. More specifically, we design a micro level space for cell structure search and a macro level cell path for better network structure modeling. Extensive experimental results in different medical image datasets verify the effectiveness of the proposed method with competitive segmentation performance, especially with minuscule neural architecture model size, i.e., 0.61M that is superior to U-Net (7.76 M) and UNet++ (9.04 M).

poster session 4 @ red 3, poster session 8 @ red 3, poster session 4, poster session 8

The task of summarization often requires a non-trivial understanding of the given text at the semantic level. In this work, we essentially incorporate the constituent structure into the single document summarization via the Graph Neural Networks to learn the semantic meaning of tokens. More specifically, we propose a novel hierarchical heterogeneous graph attention network over constituency-based parse trees for syntax-aware summarization. This approach reflects psychological findings that humans will pinpoint specific selection patterns to construct summaries hierarchically. Extensive experiments demonstrate that our model is effective for both the abstractive and extractive summarization tasks on six benchmark datasets from various domains. Moreover, further performance improvement can be obtained by virtue of state-of-the-art pre-trained models. Lastly, we test our model for the code summarization task for border impact.

poster session 3 @ red 5, poster session 8 @ red 5, poster session 3, poster session 8

Deep learning techniques for separating audio into different sound sources face several challenges. Standard architectures require training separate models for different types of audio sources. Although some universal separators employ a single model to target multiple sources, they have difficulty generalizing to unseen sources. In this paper, we propose a three-component pipeline to train a universal audio source separator from a large, but weakly-labeled dataset: AudioSet. First, we propose a transformer-based sound event detection system for processing weakly-labeled training data. Second, we devise a query-based audio separation model that leverages this data for model training. Third, we design a latent embedding processor to encode queries that specify audio targets for separation, allowing for zero-shot generalization. Our approach uses a single model for source separation of multiple sound types, and relies solely on weakly-labeled data for training. In addition, the proposed audio separator can be used in a zero-shot setting, learning to separate types of audio sources that were never seen in training. To evaluate the separation performance, we test our model on MUSDB18, while training on the disjoint AudioSet. We further verify the zero-shot performance by conducting another experiment on audio source types that are held-out from training. The model achieves comparable Source-to-Distortion Ratio (SDR) performance to current supervised models in both cases.

poster session 3 @ blue 3, poster session 8 @ blue 3, oral session 3 @ blue 3, poster session 3, poster session 8, oral session 3

Predicting and quantifying the impact of traffic accidents is necessary and critical to Intelligent Transport Systems (ITS). As a state-of-the-art technique in graph learning, current graph neural networks heavily rely on graph Fourier transform, assuming homophily among the neighborhood. However, the homophily assumption makes it challenging to characterize abrupt signals such as traffic accidents. Our paper proposes an abrupt graph wavelet network (AGWN) to model traffic accidents and predict their time durations using only one single snapshot.

poster session 3 @ blue 2, poster session 8 @ blue 2, poster session 3, poster session 8

This paper proposes a new variant of multi-play MAB model, to capture important factors of the sequential user centric selection problem arising from mobile edge computing, ridesharing applications, etc. In the proposed model, each arm is associated with discrete units of resources, each play is associate with movement costs and multiple plays can pull the same arm simultaneously. To learn the optimal action profile (an action profile prescribes the arm that each play pulls), there are two challenges: (1) the number of action profiles is large, i.e., M^K, where K and M denote the number of plays and arms respectively; (2) feedbacks on action profiles are not available, but instead feedbacks on some model parameters can be observed. To address the first challenge, we formulate a completed weighted bipartite graph to capture key factors of the offline decision problem with given model parameters. We identify the correspondence between action profiles and a special class of matchings of the graph. We also identify a dominance structure of this class of matchings. This correspondence and dominance structure enable us to design an algorithm named OffOptActPrf to locate the optimal action with a computational complexity of O(M^3K^3 ). To address the second challenge, we design an OnLinActPrf algorithm. We design estimators for model parameters and use these estimators to design a Quasi-UCB index for each action profile. The OnLinActPrf uses OffOptActPrf as a subroutine to select the action profile with the largest Quasi-UCB index. We prove that OnLinActPrf achieves a sub-linear regret. We conduct extensive experiments to validate the efficiency of OnLinActPrf.

poster session 4 @ blue 4, poster session 8 @ blue 4, oral session 4 @ blue 4, poster session 4, poster session 8, oral session 4

Configuration checking (CC) has been confirmed to alleviate the cycling problem in local search for combinatorial optimization problems (COPs). When using CC heuristics in local search for graph problems, a critical concept is the configuration of the vertices. All existing CC variants employ either 1- or 2-level neighborhoods of a vertex as its configuration. Inspired by the idea that neighborhoods with different levels should have different contributions to solving COPs, we propose the probabilistic configuration (PC), which introduces probabilities for neighborhoods at different levels to consider the impact of neighborhoods of different levels on the CC strategy. Based on the concept of PC, we first propose probabilistic configuration checking (PCC), which can be developed in an automated and lightweight favor. We then apply PCC to two classic COPs which have been shown to achieve good results by using CC, and our preliminary results confirm that PCC improves the existing algorithms because PCC alleviates the cycling problem.

poster session 4 @ blue 4, poster session 8 @ blue 4, poster session 4, poster session 8

Multiagent reinforcement learning algorithms have not been widely adopted in large scale environments with many agents as they often scale poorly with the number of agents. Using mean field theory to aggregate agents has been proposed as a solution to this problem. However, almost all previous methods in this area make a strong assumption of a centralized system where all the agents in the environment learn the same policy and are effectively indistinguishable from each other. In this paper, we relax this assumption about indistinguishable agents and propose a new mean field system known as Decentralized Mean Field Games, where each agent can be quite different from others. All agents learn independent policies in a decentralized fashion, based on their local observations. We define a theoretical solution concept for this system and provide a fixed point guarantee for a Q-learning based algorithm in this system. A practical consequence of our approach is that we can address a `chicken-and-egg' problem in empirical mean field reinforcement learning algorithms. Further, we provide Q-learning and actor-critic algorithms that use the decentralized mean field learning approach and give stronger performances compared to common baselines in this area. In our setting, agents do not need to be clones of each other and learn in a fully decentralized fashion. Hence, for the first time, we show the application of mean field learning methods in fully competitive environments, large-scale continuous action space environments, and other environments with heterogeneous agents. Importantly, we also apply the mean field method in a ride-sharing problem using a real-world dataset. We propose a decentralized solution to this problem, which is more practical than existing centralized training methods.

poster session 4 @ red 3, poster session 8 @ red 3, poster session 4, poster session 8

This paper studies a weakly supervised domain adaptation (WSDA) problem, where we only have access to the source domain with noisy labels, from which we need to transfer useful information to the unlabeled target domain. Although there have been a few studies on this problem, most of them only exploit unidirectional relationships from the source domain to the target domain. In this paper, we propose a universal paradigm called GearNet to exploit bilateral relationships between the two domains. Specifically, we take the two domains as different inputs to train two models alternately, and a symmetrical Kullback-Leibler loss is used for selectively matching the predictions of the two models in the same domain. This interactive learning schema enables implicit label noise canceling and exploit correlations between the source and target domains. Therefore, our GearNet has the great potential to boost the performance of a wide range of existing WSDA methods. Comprehensive experimental results show that the performance of existing methods can be significantly improved by equipping with our GearNet.

poster session 3 @ blue 3, poster session 8 @ blue 3, poster session 3, poster session 8

Ontology-based query answering with existential rules is well understood and implemented for positive queries, in particular conjunctive queries. For queries with negation, however, there is no agreed-upon semantics or standard implementation. This problem is unknown for simpler rule languages, such as Datalog, where it is intuitive and practical to evaluate negative queries over the least model. This fails for existential rules, which instead of a single least model have multiple universal models that may not lead to the same results for negative queries. We therefore propose universal core models as a basis for a meaningful (non-monotonic) semantics for queries with negation. Since cores are hard to compute, we identify syntactic conditions (on rules and queries) under which our core-based semantics can equivalently be obtained for other universal models, such as those produced by practical chase algorithms. Finally, we use our findings to propose a semantics for a broad class of existential rules with negation.

poster session 4 @ blue 5, poster session 8 @ blue 5, poster session 4, poster session 8

Policy gradient methods have been frequently applied to problems in control and reinforcement learning with great success, yet existing convergence analysis still relies on non-intuitive, impractical and often opaque conditions. In particular, existing rates are achieved in limited settings, under strict smoothness and bounded conditions. In this work, we establish explicit convergence rates of policy gradient methods without relying on these conditions, instead extending the convergence regime to weakly smooth policy classes with L2 integrable gradient. We provide intuitive examples to illustrate the insight behind these new conditions. We also characterize the sufficiency conditions for the ergodicity of near-linear MDPs, which represent an important class of problems. Notably, our analysis also shows that fast convergence rates are achievable for both the standard policy gradient and the natural policy gradient algorithms under these assumptions. Lastly we provide conditions and analysis for optimality of the converged policies.

poster session 1 @ blue 1, poster session 8 @ blue 1, poster session 1, poster session 8

Representation learning in dynamic graphs is a challenging problem because the topology of graph and node features vary at different time. This requires the model to be able to effectively capture both graph topology information and temporal information. Most existing works are built on recurrent neural networks (RNNs), which are used to exact temporal information of dynamic graphs, and thus they inherit the same drawbacks of RNNs. In this paper, we propose Learning to Evolve on Dynamic Graphs (LEDG) - a novel algorithm that jointly learns graph information and time information. Specifically, our approach utilizes gradient-based meta-learning to learn updating strategies that have better generalization ability than RNN on snapshots. It is model-agnostic and thus can train any message passing based graph neural network (GNN) on dynamic graphs. To enhance the representation power, we disentangle the embeddings into time embeddings and graph intrinsic embeddings. We conduct experiments on various datasets and down-stream tasks, and the experimental results validate the effectiveness of our method.

poster session 3 @ blue 2, poster session 8 @ blue 2, poster session 3, poster session 8

The paper addresses the challenge of accelerating identification of changes in risk drivers in the insurance industry. Specifically, the work presents a method to identify significant news events ("signals") from batches of news data to inform Life & Health insurance decisions. Signals are defined as events that are relevant to a tracked risk driver, widely discussed in multiple news outlets, contain novel information and affect stakeholders. The method converts unstructured data (news articles) into a sequence of keywords by employing a linguistic knowledge graph-based model. Then, for each time window, the method forms a graph with extracted keywords as nodes and draws weighted edges based on keyword co-occurrences in articles. Lastly, events are derived in an unsupervised way as graph communities and scored for the requirements of a signal: relevance, novelty and virality. The methodology is illustrated for a Life & Health topic using news articles from Dow Jones DNA proprietary data set, and assessed against baselines on a publicly available news data set. The method is implemented as an analytics engine in Early Warning System deployed at Swiss Re for the last 1.5 years to extract relevant events from live news data. We present the system's architectural design in production and discuss its use and impact.

poster session 4 @ red 6, poster session 8 @ red 6, poster session 4, poster session 8

Causality plays a central role in reasoning about observations. In many cases, it might be useful to define the conditions under which a non-deterministic program can be called an actual cause of an effect in a setting where a sequence of pro- grams are executed one after another. There can be two perspectives, one where at least one execution of the program leads to the effect, and another where all executions do so. The former captures a ``weak'' notion of causation and is more general than the latter stronger notion. In this paper, we give the definition of weak potential causes. Our analysis is performed within the situation calculus basic action theories and we consider programs formulated in the logic programming language ConGolog. Within this setting, we show how one can utilize a recently developed abstraction framework to relate causes at various levels of abstraction. In particular, we prove that under some conditions, causes defined at an abstract and a concrete level can be related with each other in a kind of commutative diagram.

poster session 4 @ blue 5, poster session 8 @ blue 5, poster session 4, poster session 8

Training deep neural networks (DNNs) for meaningful differential privacy (DP) guarantees severely degrades model utility. In this paper, we demonstrate that the architecture of DNNs has a significant impact on model utility in the context of private deep learning, whereas its effect is largely unexplored in previous studies. In light of this missing, we propose the very first framework that employs neural architecture search to automatic model design for private deep learning, dubbed as DPNAS. To integrate private learning with architecture search, a DP-aware approach is introduced for training candidate models composed on a delicately defined novel search space. We empirically certify the effectiveness of the proposed framework. The searched model DPNASNet achieves state-of-the-art privacy/utility trade-offs, e.g., for the privacy budget of $(\epsilon, \delta)=(3, 1\times10^{-5})$, our model obtains test accuracy of $98.57\%$ on MNIST, $88.09\%$ on FashionMNIST, and $68.33\%$ on CIFAR-10. Furthermore, by studying the generated architectures, we provide several intriguing findings of designing private-learning-friendly DNNs, which can shed new light on model design for deep learning with differential privacy.

poster session 3 @ blue 3, poster session 8 @ blue 3, poster session 3, poster session 8

Product allocation in retail is the process of placing products throughout a store to connect consumers with relevant products. Discovering a good allocation strategy is challenging due to the scarcity of data and the high cost of experimentation in the physical world. Some work explores Reinforcement learning (RL) as a solution, but these approaches are often limited because of the sim2real problem. Learning policies from logged trajectories of a system is a key step forward for RL in physical systems. Recent work has shown that model-based offline RL can improve the effectiveness of offline policy estimation through uncertainty-penalized exploration. However, existing work assumes a continuous state space and access to a covariance matrix of the environment dynamics, which is not possible in the discrete case. To solve this problem, we propose a Bayesian model-based technique that naturally produces probabilistic estimates of the environment dynamics via the posterior predictive distribution, which we use for uncertainty-penalized exploration. We call our approach Posterior Penalized Offline Policy Optimization (PPOPO). We show that our world model better fits historical data due to informative priors, and that PPOPO outperforms other offline techniques in simulation and against real-world data.

poster session 1 @ red 6, poster session 8 @ red 6, poster session 1, poster session 8

Existing permutation-invariant methods can be divided into two categories according to the aggregation scope, i.e. global aggregation and local one. Although the global aggregation methods, e. g., PointNet and Deep Sets, get involved in simpler structures, their performance is poorer than the local aggregation ones like PointNet++ and Point Transformer. It remains an open problem whether there exists a global aggregation method with a simple structure, competitive performance, and even much fewer parameters. In this paper, we propose a novel global aggregation permutation-invariant network based on dual MLP dot-product, called DuMLP-Pin, which is capable of being employed to extract features for set inputs, including unordered or unstructured pixel, attribute, and point cloud data sets. We strictly prove that any permutation-invariant function implemented by DuMLP-Pin can be decomposed into two or more permutation-equivariant ones in a dot-product way as the cardinality of the given input set is greater than a threshold. We also show that the DuMLP-Pin can be viewed as Deep Sets with strong constraints under certain conditions. The performance of DuMLP-Pin is evaluated on several different tasks with diverse data sets. The experimental results demonstrate that our DuMLP-Pin achieves the best results on the two classification problems for pixel sets and attribute sets. On both the point cloud classification and the part segmentation, the accuracy of DuMLP-Pin is very close to the so-far best-performing local aggregation method with only a 1-2% difference, while the number of required parameters is significantly reduced by more than 85% in classification and 69% in segmentation, respectively.

poster session 3 @ red 2, poster session 8 @ red 2, oral session 8 @ red 2, poster session 3, poster session 8, oral session 8

We present a machine-learning-guided process that can efficiently extract factor tables from unstructured rate filing documents. Our approach combines multiple deep-learning-based models that work in tandem to create structured representations of tabular data present in unstructured documents such as pdf files. This process combines CNN's to detect tables, language-based models to extract table metadata and conventional computer vision techniques to improve the accuracy of tabular data on the machine-learning side. The extracted tabular data is validated through an intuitive user interface. This process, which we call Harvest, significantly reduces the time needed to extract tabular information from PDF files, enabling analysis of such data at a speed and scale that was previously unattainable.

poster session 7 @ red 6, poster session 8 @ red 6, poster session 7, poster session 8

Quantum computers can theoretically have significant acceleration over classical computers; but, the near-future era of quantum computing is limited due to small number of qubits that are also error prone. QUILT is a framework for performing multi-class classification task designed to work effectively on current error-prone quantum computers. QUILT is evaluated with real quantum machines as well as with projected noise levels as quantum machines become more noise free. QUILT demonstrates up to 85% multi-class classification accuracy with the MNIST dataset on a five-qubit system.

poster session 4 @ blue 5, poster session 8 @ blue 5, poster session 4, poster session 8

In the past decade, automatic product description generation for e-commerce have witnessed significant advancement. As the services provided by e-commerce platforms become diverse, it is necessary to dynamically adapt the patterns of descriptions generated. The selling point of products is an important type of product description for which the length should be as short as possible while still conveying key information. In addition, this kind of product description should be eye-catching to the readers. Currently, product selling points are normally written by human experts. Thus, the creation and maintenance of these contents incur high costs. These costs can be significantly reduced if product selling points can be automatically generated by machines. In this paper, we report our experience developing and deploying the Intelligent Online Selling Point Extraction (IOSPE) system to serve the recommendation system in the JD.com e-commerce platform. Since July 2020, IOSPE has become a core service for 62 key categories of products (covering more than 4 million products). So far, it has generated more than 1.1 billion selling points, thereby significantly scaling up the selling point creation operation and saving human labour. These IOSPE generated selling points have increased the click-through rate (CTR) by 1.89% and the average duration the customers spent on the products by more than 2.03% compared to the previous practice, which are significant improvements for such a large-scale e-commerce platform.

poster session 1 @ red 6, poster session 8 @ red 6, poster session 1, poster session 8

Networked discrete dynamical systems are commonly used as formal models to study the spread of contagions over networks. Fixed points of such dynamical systems represent configurations to which systems may converge. In the case of undesirable contagions (such as epidemics, fake news and rumors) spreading over social networks (where nodes represent a population), convergence to configurations with a small number of affected nodes is highly desirable. Motivated by such considerations, we formulate a novel optimization problem for discrete dynamical systems, where the state of each node is from {0, 1} and state 1 indicates a node that has been infected by the contagion. Specifically, the goal of the problem is to find a fixed point of such a system where the number of nodes in state 1 is a minimum. We also require such a fixed point to be nontrivial (to avoid the trivial case where all nodes are in state 0). Our results are for dynamical systems where the behavior of each node is specified by a threshold function. These functions capture simple as well as complex contagions and have also been studied in game theory. We show that, unless P = NP, there is no polynomial time algorithm for approximating the solution to within the factor n^{1 - epsilon} for any constant \epsilon > 0, even when the underlying network is bipartite. We then explore several approaches to cope with the computational intractability of the problem. First, we identify several special cases (e.g., dynamical systems over directed acyclic graphs, progressive contagions where nodes cannot transition from state 1 to state 0) for which the problem can be solved efficiently. We also identify a version that is fixed parameter tractable with respect to a structural parameter associated with the problem. Next, we develop an integer linear programming (ILP) formulation for the problem so that existing solvers such as Gurobi can be used to solve the problem in practice for networks of reasonable size. For larger networks, we propose a general heuristic framework for the problem. We present extensive experimental results using real-world networks to demonstrate the effectiveness of the proposed heuristic framework.

poster session 4 @ red 3, poster session 8 @ red 3, oral session 4 @ red 3, poster session 4, poster session 8, oral session 4

In this paper, we propose the first self-distillation framework for general object detection, termed LGD (Label-Guided self-Distillation). Previous studies rely on a strong pretrained teacher to provide instructive knowledge that could be unavailable in real-world scenarios. Instead, we generate an instructive knowledge by inter-and-intra relation modeling among objects, requiring only student representations and regular labels. Concretely, our framework involves sparse label-appearance encoding, inter-object relation adaptation and intra-object knowledge mapping to obtain the instructive knowledge. They jointly form an implicit teacher at training phase, dynamically dependent on labels and evolving student representations. Modules in LGD are trained end-to-end with student detector and are discarded in inference. Experimentally, LGD obtains decent results on various detectors, datasets, and extensive tasks like instance segmentation. For example in MS-COCO dataset, LGD improves RetinaNet with ResNet-50 under 2x single-scale training from 36.2% to 39.0% mAP (+ 2.8%). It boosts much stronger detectors like FCOS with ResNeXt-101 DCN v2 under 2x multi-scale training from 46.1% to 47.9% (+ 1.8%).

poster session 3 @ red 2, poster session 8 @ red 2, poster session 3, poster session 8

Knowledge Distillation (KD), which is an effective model compression and acceleration technique, has been successfully applied to graph neural networks (GNNs) recently. Existing approaches utilize a single GNN model as the teacher to distill knowledge. However, we notice that GNN models with different number of layers demonstrate different classification abilities on nodes with different degrees. On the one hand, for nodes with high degrees, their local structures are dense and complex, hence more message passing is needed. Therefore, GNN models with more layers perform better. On the other hand, for nodes with low degrees, whose local structures are relatively sparse and simple, the repeated message passing can easily lead to over-smoothing. Thus, GNN models with less layers are more suitable. However, existing single-teacher GNN knowledge distillation approaches which are based on a single GNN model, are sub-optimal. To this end, we propose a novel approach to distill multi-scale knowledge, which learns from multiple GNN teacher models with different number of layers to capture the topological semantic at different scales. Instead of learning from the teacher models equally, the proposed method automatically assigns proper weights for each teacher model via an attention mechanism which enables the student to select teachers for different local structures. Extensive experiments are conducted to evaluate the proposed method on four public datasets. The experimental results demonstrate the superiority of our proposed method over state-of-the-art methods.

poster session 3 @ blue 3, poster session 8 @ blue 3, oral session 8 @ blue 3, poster session 3, poster session 8, oral session 8

Federated learning (FL) is a popular technique to train machine learning (ML) models with decentralized data. Extensive works have studied the performance of the global model; however, it is still unclear how the training process affects the final test accuracy. Exacerbating this problem is the fact that FL executions differ significantly from traditional ML with heterogeneous data characteristics across clients, involving more hyperparameters. In this work, we show that the final test accuracy of FL is dramatically affected by the early phase of the training process, i.e., FL exhibits critical learning periods, in which small gradient errors can have irrecoverable impact on the final test accuracy. To further explain this phenomenon, we generalize the trace of the Fisher Information Matrix (FIM) to FL and define a new notation called FedFIM, a quantity reflecting the local curvature of each clients from the beginning of the training in FL. Our findings suggest that the initial learning phase plays a critical role in understanding the FL performance. This is in contrast to many existing works which generally do not connect the final accuracy of FL to the early phase training. Finally, seizing critical learning periods in FL is of independent interest and could be useful for other problems such as the choices of hyperparameters such as the number of client selected per round, batch size, and more, so as to improve the performance of FL training and testing.

poster session 3 @ blue 3, poster session 8 @ blue 3, poster session 3, poster session 8

Reinforcement learning (RL) has demonstrated its superiority in solving sequential decision-making problems. However, heavy dependence on immediate reward feedback impedes the wide application of RL. On the other hand, imitation learning (IL) tackles RL without relying on environmental supervision by leveraging external demonstrations. In practice, however, collecting sufficient expert demonstrations can be prohibitively expensive, yet the quality of demonstrations typically limits the performance of the learning policy. To address a practical scenario, in this work, we propose Self-Adaptive Imitation Learning (SAIL), which, provided with a few demonstrations from a sub-optimal teacher, can perform well in RL tasks with extremely delayed rewards, where the only reward feedback is trajectory-wise ranking. SAIL bridges the advantages of IL and RL by interactively exploiting the demonstrations to catch up with the teacher and exploring the environment to yield demonstrations that surpass the teacher. Extensive empirical results show that not only does SAIL significantly improve the sample efficiency, but it also leads to higher asymptotic performance across different continuous control tasks, compared with the state-of-the-art.

poster session 1 @ blue 1, poster session 8 @ blue 1, poster session 1, poster session 8

For Boolean classifiers, the {\em complete reason} behind a decision is a Boolean formula that characterizes why the decision was made. This recently introduced notion has a number of applications, which include generating explanations, detecting decision bias and evaluating counterfactual queries. Prime {\em implicants} of the complete reason are known as {\em sufficient reasons} for the decision and they correspond to what is known as PI explanations and abductive explanations. In this paper, we refer to the prime {\em implicates} of a complete reason as {\em necessary reasons} for the decision. We justify this terminology semantically and show that necessary reasons correspond to what is known as contrastive explanations. We also study the computation of complete reasons for multi-class decision trees and graphs with nominal and numeric features for which we derive efficient, closed-form complete reasons. We further investigate the computation of shortest necessary and sufficient reasons for a broad class complete reasons, which include the derived closed forms and the complete reasons for Sentential Decision Diagrams (SDDs). We provide an algorithm which can enumerate their shortest necessary reasons in output polynomial time. Enumerating shortest sufficient reasons for this class of complete reasons is hard even for a single reason. For this problem, we provide an algorithm that appears to be quite efficient as we show empirically.

poster session 4 @ blue 5, poster session 8 @ blue 5, poster session 4, poster session 8

Interactive AI (IAI) systems are increasingly popular as the human-centered AI design paradigm is gaining strong traction. However, evaluating IAI systems, a key step in building such systems, is particularly challenging, as their output highly depends on the performed user actions. Developers often have to rely on limited and mostly qualitative data from ad-hoc user testing to assess and improve their systems. In this paper, we present InteractEva; a systematic evaluation framework for IAI systems. We also describe how we have applied InteractEva to evaluate a commercial IAI system, leading to both quality improvements and better data-driven design decisions.

poster session 4 @ red 6, poster session 8 @ red 6, poster session 4, poster session 8

Model-free reinforcement learning (RL) algorithms are known to be memory and computation-efficient compared with model-based approaches. This paper presents a model-free RL algorithm for infinite-horizon average-reward Constrained Markov Decision Processes (CMDPs), which achieves sub-linear regret and zero constraint violation. To the best of our knowledge, this is the first {\em model-free} algorithm for general CMDPs in the infinite-horizon average-reward setting with provable guarantees.

poster session 1 @ blue 1, poster session 8 @ blue 1, poster session 1, poster session 8

Pattern mining is an essential part of knowledge discovery and data analytics. It is a powerful paradigm, especially when combined with constraint reasoning. In this paper, we present Seq2Pat, a constraint-based sequential pattern mining tool with a high-level declarative user interface. The library finds patterns that frequently occur in large sequence databases subject to constraints. We highlight key benefits that are desirable, especially in industrial settings where scalability, explainability, rapid experimentation, reusability, and reproducibility are of great interest. We then showcase an automated feature extraction process powered by Seq2Pat to discover high-level insights and boost downstream machine learning models for customer intent prediction.

poster session 4 @ red 6, poster session 8 @ red 6, poster session 4, poster session 8

Multi-modal person Re-ID introduces more complementary information to assist the traditional Re-ID task. Existing multi-modal methods ignore the importance of modality-specific information in the feature fusion stage. To this end, we propose a novel method to boost modality-specific representations for multi-modal person Re-ID: Interact, Embed, and EnlargE (IEEE). First, we propose a cross-modal interacting module to exchange useful information between different modalities in the feature extraction phase. Second, we propose a relation-based embedding module to enhance the richness of feature descriptors by embedding the global feature into the fine-grained local information. Finally, we propose multi-modal margin loss to force the network to learn modality-specific information for each modality by enlarging the intra-class discrepancy. Superior performance on multi-modal Re-ID dataset RGBNT201 and three constructed Re-ID datasets validate the effectiveness of the proposed method comparing with the state-of-the-art approaches.

poster session 3 @ red 1, poster session 8 @ red 1, poster session 3, poster session 8

We study regret minimization in finite horizon tabular Markov decision processes (MDPs) under the constraints of differential privacy (DP). This is motivated by the widespread applications of reinforcement learning (RL) in real-world sequential decision making problems, where protecting users' sensitive and private information is becoming paramount. We consider two variants of DP -- joint DP (JDP), where a centralized agent is responsible for protecting users' sensitive data and local DP (LDP), where information needs to be protected directly on the user side. We first propose two general frameworks -- one for policy optimization and another for value iteration -- for designing private, optimistic RL algorithms. We then instantiate these frameworks with suitable privacy mechanisms to satisfy JDP and LDP requirements, and simultaneously obtain sublinear regret guarantees. The regret bounds show that under JDP, the cost of privacy is only a lower order additive term, while for a stronger privacy protection under LDP, the cost suffered is multiplicative. Finally, the regret bounds are obtained by a unified analysis, which, we believe, can be extended beyond tabular MDPs.

poster session 1 @ blue 1, poster session 8 @ blue 1, oral session 1 @ blue 1, poster session 1, poster session 8, oral session 1

Speech enhancement aims at recovering a clean speech from a noisy input, which can be classified into single speech enhancement and personalized speech enhancement. Personalized speech enhancement usually utilizes the speaker identity extracted from the noisy speech itself (or a clean reference speech) as a global embedding to guide the enhancement process. Different from them, we observe that the speeches of the same speaker are correlated in terms of frame-level short-time Fourier Transform (STFT) spectrogram. Therefore, we propose reference-based speech enhancement via a feature alignment and fusion network (FAF-Net). Given a noisy speech and a clean reference speech spoken by the same speaker, we first propose a feature level alignment strategy to warp the clean reference with the noisy speech in frame level. Then, we fuse the reference feature with the noisy feature via a similarity-based fusion strategy. Finally, the fused features are skipped connected to the decoder, which generates the enhanced results. Experimental results demonstrate that the performance of the proposed FAF-Net is close to state-of-the-art speech enhancement methods on both DNS and Voice Bank+DEMAND datasets. Our code is available at https://github.com/HieDean/FAF-Net.

poster session 1 @ red 4, poster session 8 @ red 4, poster session 1, poster session 8

Answer set programming (ASP) is a popular declarative programming paradigm with a wide range of applications in artificial intelligence. Oftentimes, when modeling an AI problem with ASP, and in particular when we are interested beyond simple search for optimal solutions, an actual solution, differences between solutions, or number of solutions of the ASP program matter. For example, when a user aims to identify a specific answer set according to her needs, or requires the total number of diverging solutions to comprehend probabilistic applications such as reasoning in medical domains. Then, there are only certain problem specific and handcrafted encoding techniques available to navigate the solution space of ASP programs, which is oftentimes not enough. In this paper, we propose a formal and general framework for interactive navigation toward desired subsets of answer sets analogous to faceted browsing. Our approach enables the user to explore the solution space by consciously zooming in or out of sub-spaces of solutions at a certain configurable pace. We illustrate that weighted faceted navigation is computationally hard. Finally, we provide an implementation of our approach that demonstrates the feasibility of our framework for incomprehensible solution spaces.

poster session 4 @ blue 5, poster session 8 @ blue 5, oral session 4 @ blue 5, poster session 4, poster session 8, oral session 4

There has been significant recent interest in quantum neural networks (QNNs), along with their applications in diverse domains. Current solutions for QNNs pose significant challenges concerning their scalability, ensuring that the postulates of quantum mechanics are satisfied and that the networks are physically realizable. The exponential state space of QNNs poses significant challenges for the scalability of training procedures. The no-cloning principle prohibits making multiple copies of training samples, and the measurement postulates lead to non-deterministic loss functions. Consequently, the physical realizability and efficiency of existing approaches that rely on repeated measurement of several copies of each sample for training QNNs are unclear. This paper presents a new model for QNNs that relies on band-limited Fourier expansions of transfer functions of quantum perceptrons (QPs) to design scalable training procedures. We augment this training procedure with techniques that eliminate the need for sample replication. We show that our novel stochastic gradient descent (SGD) procedure converges to the true minima in expectation, even in the presence of non-determinism due to quantum measurement. Our solution has a number of important benefits: (i) using QPs with concentrated Fourier power spectrum, we show that the training procedure for QNNs can be made scalable; (ii) eliminates the need for resampling, thus staying consistent with the no-cloning rule; and (iii) enhanced data efficiency for the overall training process since each data sample is processed once per epoch. We present a detailed theoretical foundation for our models and methods' scalability, accuracy, and data effectiveness. Finally, we demonstrate the utility of our approach through a series of numerical experiments.

poster session 3 @ blue 3, poster session 8 @ blue 3, oral session 3 @ blue 3, poster session 3, poster session 8, oral session 3

Emotion recognition in conversation (ERC) aims to detect the emotion label for each utterance. Motivated by recent studies which have proven that feeding training examples in a meaningful order rather than considering them randomly can boost the performance of models, we propose an ERC-oriented hybrid curriculum learning framework. Our framework consists of two curricula: (1) conversation-level curriculum (CC); and (2) utterance-level curriculum (UC). In CC, we construct a difficulty measurer based on ``emotion shift'' frequency within a conversation, then the conversations are scheduled in an ``easy to hard" schema according to the difficulty score returned by the difficulty measurer. For UC, it is implemented from an emotion-similarity perspective, which progressively strengthens the model’s ability in identifying the similar emotion labels. With the proposed model-agnostic hybrid curriculum learning strategy, we observe significant performance boosts over a wide range of existing ERC models and we are able to achieve SOTA results on four public ERC datasets.

poster session 1 @ red 4, poster session 8 @ red 4, poster session 1, poster session 8

Unsupervised domain adaptive person re-identification aims at learning on an unlabeled target domain with only labeled data in source domain. Currently, the state-of-the-arts usually solve this problem by pseudo-label-based clustering and fine-tuning in target domain. However, the reason behind the noises of pseudo labels is not sufficiently explored, especially for the popular multi-branch models. We argue that the consistency between different feature spaces is the key to the pseudo labels’ quality. Then a \textbf{SE}lf-\textbf{C}onsistent pseudo label \textbf{R}efin\textbf{E}men\textbf{T} method, termed as SECRET, is proposed to improve consistency by mutually refining the pseudo labels generated from different feature spaces. The proposed SECRET gradually encourages the improvement of pseudo labels’ quality during training process, which further leads to better cross-domain Re-ID performance. Extensive experiments on benchmark datasets show the superiority of our method. Specifically, our method outperforms the state-of-the-arts by $\mathbf{6.3\%}$ in terms of mAP on the challenging dataset MSMT17. In the purely unsupervised setting, our method also surpasses existing works by a large margin. Code is available at \url{https://github.com/LunarShen/SECRET}.

poster session 3 @ red 1, poster session 8 @ red 1, poster session 3, poster session 8

Occluded person re-identification is a challenging task as human body parts could be occluded by some obstacles (e.g. trees, cars, and pedestrians) in certain scenes. Some existing pose-guided methods solve this problem by aligning body parts according to graph matching, but these graph-based methods are not intuitive and complicated. Therefore, we propose a transformer-based Pose-guided Feature Disentangling (PFD) method by utilizing pose information to clearly disentangle semantic components (e.g. human body or joint parts) and selectively match non-occluded parts correspondingly. First, Vision Transformer (ViT) is used to extract the patch features with its strong capability. Second, to preliminarily disentangle the pose information from patch information, the matching and distributing mechanism is leveraged in Pose-guided Feature Aggregation (PFA) module. Third, a set of learnable semantic views are introduced in transformer decoder to implicitly enhance the disentangled body part features. However, those semantic views are not guaranteed to be related to the body without additional supervision. Therefore, Pose-View Matching (PVM) module is proposed to explicitly match visible body parts and automatically separate occlusion features. Fourth, to better prevent the interference of occlusions, we design a Pose-guided Push Loss to emphasize the features of visible body parts. Extensive experiments over five challenging datasets for two tasks (occluded and holistic Re-ID) demonstrate that our proposed PFD is superior promising, which performs favorably against state-of-the-art methods. Code is available at https://github.com/WangTaoAs/PFD_Net

poster session 3 @ red 1, poster session 8 @ red 1, poster session 3, poster session 8

Contrastive learning models have achieved great success in unsupervised visual representation learning, which maximizes the similarities between feature representations of different views of the same image, while minimizing the similarities between feature representations of views of different images. In text summarization, the output summary is a shorter form of the input document and they have similar meanings. In this paper, we propose a contrastive learning model for supervised abstractive text summarization, where we view a document, its gold summary and its model generated summaries as different views of the same mean representation and maximize the similarities between them during training. We improve over a strong sequence-to-sequence text generation model (i.e., BART) on three different summarization datasets. Human evaluation also shows that our model achieves better faithfulness ratings compared to its counterpart without contrastive objectives.

poster session 1 @ red 4, poster session 8 @ red 4, poster session 1, poster session 8

Graph matching aims to establish correspondences between vertices of graphs such that both the node and edge attributes agree. Various learning-based methods were recently proposed for finding correspondences between image key points based on deep graph matching formulations. While these approaches mainly focus on learning node and edge attributes, they completely ignore the 3D geometry of the underlying 3D objects depicted in the 2D images. We fill this gap by proposing a trainable framework that takes advantage of graph neural networks for learning a deformable 3D geometry model from inhomogeneous image collections, i.e. a set of images that depict different instances of objects from the same category. Experimentally we demonstrate that our method outperforms recent learning-based approaches for graph matching considering both accuracy and cycle-consistency error, while we in addition obtain the underlying 3D geometry of the objects depicted in the 2D images.

poster session 4 @ red 3, poster session 8 @ red 3, poster session 4, poster session 8

Recurrent neural networks have proven effective in modeling sequential user feedbacks for recommender systems. However, they usually focus solely on item relevance and fail to effectively explore diverse items for users, therefore harming the system performance in the long run. To address this problem, we propose a new type of recurrent neural networks, dubbed recurrent exploration networks (REN), to jointly perform representation learning and effective exploration in the latent space. REN tries to balance relevance and exploration while taking into account the uncertainty in the representations. Our theoretical analysis shows that REN can preserve the rate-optimal sublinear regret even when there exists uncertainty in the learned representations. Our empirical study demonstrates that REN can achieve satisfactory long-term rewards on both synthetic and real-world recommendation datasets, outperforming state-of-the-art models.

poster session 4 @ blue 4, poster session 8 @ blue 4, poster session 4, poster session 8

Existing research in the field of automated negotiation considers a negotiation architecture in which some of the negotiation components are designed separately by reinforcement learning (RL), but comprehensive negotiation strategy design has not been achieved. In this study, we formulated an RL model based on a Markov decision process (MDP) for bilateral multi-issue negotiations. We propose a versatile negotiating agent that can effectively learn various negotiation strategies and domains through comprehensive strategies using deep RL. We show that the proposed method can achieve the same or better utility than existing negotiation agents.

poster session 4 @ blue 5, poster session 8 @ blue 5, poster session 4, poster session 8

Traffic forecasting is one of the most popular spatio-temporal tasks in the field of machine learning. A prevalent approach in the field is to combine graph convolutional networks and recurrent neural networks for the spatio-temporal processing. There has been fierce competition and many novel methods have been proposed. In this paper, we present the method of spatio-temporal graph neural controlled differential equation (STG-NCDE). Neural controlled differential equations (NCDEs) are a breakthrough concept for processing sequential data. We extend the concept and design two NCDEs: one for the temporal processing and the other for the spatial processing. After that, we combine them into a single framework. We conduct experiments with 6 benchmark datasets and 20 baselines. STG-NCDE shows the best accuracy in all cases, outperforming all those 20 baselines by non-trivial margins.

poster session 3 @ blue 2, poster session 8 @ blue 2, oral session 8 @ blue 2, poster session 3, poster session 8, oral session 8

Most prior work on online matching problems has been with the flexibility of keeping some vertices unmatched. We study three related online matching problems with the constraint of matching every vertex, i.e., with no rejections. We adopt a model in which vertices arrive in a uniformly random order and the edge-weights are arbitrary positive numbers. For the capacitated online bipartite matching problem in which the vertices of one side of the graph are offline and those of the other side arrive online, we give a 4.62-competitive algorithm when the capacity of each offline vertex is 2. For the online general (non-bipartite) matching problem, where all vertices arrive online, we give a 3.34-competitive algorithm. We also study the online roommate matching problem (Huzhang et al. 2017) in which each room (offline vertex) holds 2 persons (online vertices). Persons derive non-negative additive utilities from their room as well as roommate. In this model, with the goal of maximizing the sum of utilities of all persons, we give a 7.96-competitive algorithm. This is an improvement over the 24.72 approximation factor in (Huzhang et al. 2017).

poster session 5 @ blue 6, poster session 8 @ blue 6, poster session 5, poster session 8

Data-free knowledge distillation (DFKD) has recently been attracting increasing attention from research communities, attributed to its capability of compressing a model only using synthetic data. Despite the encouraging results achieved, state-of-the-art DFKD methods still suffer from the inefficiency of data synthesis, making the data-free training process extremely time-consuming and thus inapplicable for large-scale tasks. In this work, we introduce an efficacious scheme, termed as FastDFKD, that allows us to accelerate DFKD by a factor of orders of magnitude. At the heart of our approach is a novel strategy to reuse the shared common features in training data so as to synthesize different data instances. Unlike prior methods that optimize a set of data independently, we propose to learn a meta-synthesizer that seeks common features as the initialization for the fast data synthesis. As a result, FastDFKD achieves data synthesis within only a few steps, significantly enhancing the efficiency of data-free training. Experiments over CIFAR, NYUv2 and ImageNet demonstrate that the proposed FastDFKD achieves 10$\times$ and even 100$\times$ acceleration while preserving performances on par with the state of the art.

poster session 3 @ blue 3, poster session 8 @ blue 3, poster session 3, poster session 8

We formalize a voting model for plurality elections that combines Iterative Voting and Calculus of Voting. Each iteration, autonomous agents simultaneously maximize the utility they expect from candidates. Agents are aware of neither other individuals’ preferences or choices, nor of the distribution preferences come from. They know only of candidates’ current expected vote shares (e.g. from a poll, network, etc) and with that calculate expected rewards from each candidate, pondering the probability that voting for each would alter the final outcome. We define the general form of those pivotal probabilities, then we derive both efficient exact calculations as well as heuristics. Lastly, we both prove formally that the model converges with asymptotically large electorates and show empirically (via simulations) that it nearly always converges even with very few agents.

poster session 5 @ blue 6, poster session 8 @ blue 6, poster session 5, poster session 8

In bi-objective search, we are given a graph in which each directed arc is associated with a pair of non-negative weights, and the objective is to find the Pareto-optimal solution set. Unfortunately, in many practical settings, this set is too large, and therefore its computation is very time-consuming. In addition, even though bi-objective search algorithms generate the Pareto set incrementally, they do so exhaustively. This means that early during search the solution set covers is not diverse, being concentrated in a small region of the solution set. To address this issue, we present a new approach to subset approximation of the solution set, that can be used as the basis for an anytime bi-objective search algorithm. Our approach transforms the given task into a target bi-objective search task using two real parameters. For each particular parameter setting, the solutions to the target task is a subset of the solution set of the original task. Depending on the parameters used, the solution set of the target task may be computed very quickly. This allows us to obtain, in challenging road map benchmarks, a rich variety of solutions in times that may be orders of magnitude smaller than the time needed to compute the solution set. We show that by running the algorithm with an appropriate sequence of parameters, we obtain a growing sequence of solutions that converges to the full solution set. We prove that our approach is correct and that Bi-Objective A* prunes at least as many nodes when run over the target task.

poster session 4 @ blue 4, poster session 8 @ blue 4, oral session 4 @ blue 4, poster session 4, poster session 8, oral session 4

Product copywriting is a critical component of e-commerce recommendation platforms. It aims to attract users' interest and improve user experience by highlighting product characteristics with textual descriptions. In this paper, we report our experience deploying the proposed Automatic Product Copywriting Generation (APCG) system into the JD.com e-commerce product recommendation platform. It consists of two main components: 1) natural language generation, which is built from a transformer-pointer network and a pre-trained sequence-to-sequence model based on millions of training data from our in-house platform; and 2) copywriting quality control, which is based on both automatic evaluation and human screening. For selected domains, the models are trained and updated daily with the updated training data. In addition, the model is also used as a real-time writing assistant tool on our live broadcast platform. The APCG system has been deployed in JD.com since Feb 2021. By Sep 2021, it has generated 2.53 million product descriptions, and improved the overall averaged click-through rate (CTR) and the Conversion Rate (CVR) by 4.22% and 3.61%, compared to baselines, respectively on a year-on-year basis. The accumulated Gross Merchandise Volume (GMV) made by our system is improved by 213.42%, compared to the number in Feb 2021.

poster session 1 @ red 6, poster session 8 @ red 6, poster session 1, poster session 8

Human conversations are guided by short-term and long-term goals. We study how to plan short-term goal sequences as coherently as humans do and naturally direct them to an assigned long-term goal in open-domain conversations. Goal sequences are a series of knowledge graph (KG) entity-relation connections generated by KG walkers that traverse through the KG. The existing recurrent and graph attention based KG walkers either insufficiently utilize the conversation states or lack global guidance. In our work, a hierarchical model learns goal planning in a hierarchical learning framework. We present HiTKG, a hierarchical transformer-based graph walker that leverages multiscale inputs to make precise and flexible predictions on KG paths. Furthermore, we propose a two-hierarchy learning framework that employs two stages to learn both turn-level (short-term) and global-level (long-term) conversation goals. Specifically, at the first stage, HiTKG is trained in a supervised fashion to learn how to plan turn-level goal sequences; at the second stage, HiTKG tries to naturally approach the assigned global goal via reinforcement learning. In addition, we propose MetaPath as the backbone method for KG path representation to exploit the entity and relation information concurrently. We further propose Multi-source Decoding Inputs and Output-level Length Head to improve the decoding controllability. Our experiments show that HiTKG achieves a significant improvement in the performance of turn-level goal learning compared with state-of-the-art baselines. Additionally, both automatic and human evaluation prove the effectiveness of the two-hierarchy learning framework for both short-term and long-term goal planning.

poster session 3 @ red 5, poster session 8 @ red 5, oral session 3 @ red 5, poster session 3, poster session 8, oral session 3

We present a multi-objective meta-search procedure that constructs candidate algorithms for state-space search puzzles like Rubik's cube. The candidate algorithms take the form of macro databases, i.e., rule tables that specify sequences of actions to perform in different states. Rules are repeatedly applied until the puzzle is solved. The objectives favor candidates that are god-like (solving the puzzle in fewer steps) and folk-like (having fewer rules in the macro database). We build each candidate with a non-deterministic rule table construction, and then optimize over the non-deterministic choice points to find candidates near the Pareto-optimal trades-offs between godliness and folksiness. We prove that the rule table construction is correct: it always terminates and solves every state at termination. This is verified empirically on the full 2x2x2 "pocket" cube, where correct (but unoptimized) constructions take under one hour and the total number of rules is less than 10% the number of possible states. We also empirically assess the multi-objective optimization on restricted variants of the cube with up to 29K possible states, showing relative improvements in the objectives between 14-20%. Avenues for scaling up the method in future work are discussed.

poster session 4 @ blue 4, poster session 8 @ blue 4, poster session 4, poster session 8

Given a query graph and a database of corpus graphs, a graph retrieval system aims to deliver the most relevant corpus graphs. Graph retrieval based on subgraph matching has a wide variety of applications, e.g., molecular fingerprint detection, circuit design, and question answering. In this setup, a corpus graph is relevant to a query graph, if this query graph is a subgraph of the corpus graph. Existing neural graph retrieval models compare the node or graph embeddings of the query-corpus pairs to compute the relevance scores between them. However, such models may not provide edge consistency between query corpus pairs. Moreover, they predominantly use symmetric relevance scores, which are not appropriate in the context of subgraph matching, since the underlying relevance score in subgraph search should be measured using the partial order induced by subgraph-supergraph relationship. Consequently, they show poor retrieval performance in the context of subgraph matching. In response, we propose ISONET, an interpretable neural edge alignment model, which is better able to learn the edge consistent mapping necessary for subgraph matching. Furthermore, we design a novel scoring mechanism which enforces an asymmetric relevance score specifically for subgraph matching. ISONET’s design enables it to identify the underlying subgraph in corpus graph, which is relevant to the given query graph. Our experiments on diverse datasets show that ISONET outperforms existing baseline models for graph retrieval. Additionally, ISONET can provide the interpretable alignments between query-corpus graph pairs during inference, despite being trained only using binary relevance labels during training, without any ground truth information about node or edge alignments.

poster session 3 @ blue 2, poster session 8 @ blue 2, poster session 3, poster session 8

The Latin square completion (LSC) problem is an important NP-complete problem with numerous applications. Given its theoretical and practical importance, several algorithms are designed for solving the LSC problem. In this work, to further improve the performance, a fast local search algorithm is developed based on three main ideas. Firstly, a reduction reasoning technique is used to reduce the scale of search space. Secondly, we propose a novel conflict value selection heuristic, which considers the history conflicting information of vertices as a selection criterion when more than one vertex have equal values on the primary scoring function. Thirdly, during the search phase, we record previous history search information and then make use of these information to restart the candidate solution. Experimental results show that our proposed algorithm significantly outperforms the state-of-the-art heuristic algorithms on almost all instances in terms of success rate and run time.

poster session 4 @ blue 4, poster session 8 @ blue 4, poster session 4, poster session 8

Recent studies in the hate and counter hate domain have provided the grounds for investigating how to detect this pervasive content in social media. These studies mostly work with synthetic replies to hateful content written by annotators on demand rather than replies written by real users. We argue that working with naturally occurring replies to hateful content is key to study the problem. Building on this motivation, we create a corpus of 5,652 hateful tweets and replies. We analyze their fine-grained relationships by indicating whether the reply (a) is hate or counter hate speech, (b) provides a justification, (c) attacks the author of the tweet, and (d) adds additional hate. We also present linguistic insights into the language people use depending on these fine-grained relationships. Experimental results show improvements (a) taking into account the hateful tweet in addition to the reply and (b) pretraining with related tasks.

poster session 1 @ red 4, poster session 8 @ red 4, oral session 8 @ red 4, poster session 1, poster session 8, oral session 8

Sarcasm is a pervading linguistic phenomenon and highly challenging to explain due to its subjectivity, lack of context and deeply-felt opinion. In the multimodal setup, sarcasm is conveyed through the incongruity between the text and visual entities. Although recent approaches deal with sarcasm as a classification problem, it is unclear why an online post is identified as sarcastic. Without proper explanation, end users may not be able to perceive the underlying sense of irony. In this paper, we propose a novel problem -- Multimodal Sarcasm Explanation (MuSE) -- given a multimodal sarcastic post containing an image and a caption, we aim to generate a natural language explanation to reveal the intended sarcasm. To this end, we develop MORE, a new dataset with explanation of 3510 sarcastic multimodal posts. Each explanation is a natural language (English) sentence describing the hidden irony. We benchmark MORE by employing a multimodal Transformer-based architecture. It incorporates a cross-modal attention in the Transformer's encoder which attends to the distinguishing features between the two modalities. Subsequently, a BART-based auto-regressive decoder is used as the generator. Empirical results demonstrate convincing results over various baselines (adopted for MuSE) across five evaluation metrics. We also conduct human evaluation on predictions and obtain Fleiss' Kappa score of 0.4 as a fair agreement among 25 evaluators.

poster session 1 @ red 4, poster session 8 @ red 4, poster session 1, poster session 8

This paper presents SimCTC, a simple contrastive learning (CL) framework that greatly advances the state-of-the-art text clustering models. In SimCTC, a pre-trained BERT model first maps the input sequence to the representation space, which is then followed by three different loss function heads: Clustering head, Instance-CL head and Cluster-CL head. Experimental results on multiple benchmark datasets demonstrate that SimCTC remarkably outperforms 6 competitive text clustering methods with 1%-6% improvement on Accuracy (ACC) and 1%-4% improvement on Normalized Mutual Information (NMI). Moreover, our results also show that the clustering performance can be further improved by setting an appropriate number of clusters in the cluster-level objective.

poster session 4 @ blue 5, poster session 8 @ blue 5, poster session 4, poster session 8

Item representation learning is crucial for search and recommendation tasks in e-commerce. In e-commerce, the instances (e.g., items, users) in different domains are always related. Such instance relationship across domains contains useful local information for transfer learning. However, existing transfer learning based approaches did not leverage this knowledge. In this paper, we report on our experience designing and deploying Prior-Guided Transfer Learning (PGTL) to bridge this gap. It utilizes the instance relationship across domains to extract prior knowledge for the target domain and leverages it to guide the fine-grained transfer learning for e-commerce item representation learning tasks. Rather than directly transferring knowledge from the source domain to the target domain, the prior knowledge can serve as a bridge to link both domains and enhance knowledge transfer, especially when the domain distribution discrepancy is large. Since its deployment on the Taiwanese portal of Taobao in Aug 2020, PGTL has significantly improved the item exposure rate and item click-through rate compared to previous approaches

poster session 1 @ red 6, poster session 8 @ red 6, poster session 1, poster session 8

Tremendous progress has been made in sequential processing with the recent advances in recurrent neural networks. However, recurrent architectures face the challenge of exploding/vanishing gradients during training, and require significant computational resources to execute back-propagation through time. Moreover, large models are typically needed for executing complex sequential tasks. To address these challenges, we propose a novel neuron model that has cosine activation with a time varying component for sequential processing. The proposed neuron provides an efficient building block for projecting sequential inputs into spectral domain, which helps to retain long-term dependencies with minimal extra model parameters and computation. A new type of recurrent network architecture, named Oscillatory Fourier Neural Network, based on the proposed neuron is presented and applied to various types of sequential tasks. We demonstrate that recurrent neural network with the proposed neuron model is mathematically equivalent to a simplified form of discrete Fourier transform applied onto periodical activation. In particular, the computationally intensive back-propagation through time in training is eliminated, leading to faster training while achieving the state of the art inference accuracy in a diverse group of sequential tasks. For instance, applying the proposed model to sentiment analysis on IMDB review dataset reaches 89.4% test accuracy within 5 epochs, accompanied by over 35x reduction in the model size compared to LSTM. The proposed novel RNN architecture is well poised for intelligent sequential processing in resource constrained hardware.

poster session 1 @ red 4, poster session 8 @ red 4, poster session 1, poster session 8

Accounts Payable (AP) is a resource-intensive business process in large enterprises for paying vendors within contractual payment deadlines for goods and services procured from them. There are multiple verifications before payment to the supplier/vendor. After the validations, the invoice flows through several steps such as vendor identification, line-item matching for Purchase order (PO) based invoices, Accounting Code identification for Non- Purchase order (Non-PO) based invoices, tax code identification, etc. Currently, each of these steps is mostly manual and cumbersome making it labor-intensive, error-prone, and requiring constant training of agents. Automatically processing these invoices for payment without any manual intervention is quite difficult. To tackle this challenge, we have developed an automated end-to-end invoice processing system using AI-based modules for multiple steps of the invoice processing pipeline. It can be configured to an individual client’s requirements with minimal effort. Currently, the system is deployed in production for two clients. It has successfully processed around ~80k invoices out of which 76% invoices were processed with low or no manual intervention.

poster session 4 @ red 6, poster session 8 @ red 6, poster session 4, poster session 8

Although convolution neural network based stereo matching architectures have made impressive achievements, there are still some limitations: 1) Convolutional Feature (CF) tends to capture appearance information, which is inadequate for accurate matching. 2) Due to the static filters, current convolution based disparity refinement modules often produce over-smooth results. In this paper, we present two schemes to address these issues, where some traditional wisdoms are integrated. Firstly, we introduce a pairwise feature for deep stereo matching networks, named LSP (Local Similarity Pattern). Through explicitly revealing the neighbor relationships, LSP contains rich structural information, which can be leveraged to aid CF for more discriminative feature description. Secondly, we design a dynamic self-reassembling refinement strategy and apply it to the cost distribution and the disparity map respectively. The former could be equipped with the unimodal distribution constraint to alleviate the over-smoothing problem, and the latter is more practical. The effectiveness of the proposed methods is demonstrated via incorporating them into two well-known basic architectures, GwcNet and GANet-deep. Experimental results on the SceneFlow and KITTI benchmarks show that our modules significantly improve the performance of the model.

poster session 3 @ red 2, poster session 8 @ red 2, poster session 3, poster session 8

Inspired by the extensive success of deep learning, graph neural networks (GNNs) have been proposed to learn expressive node representations and demonstrated promising performance in various graph learning tasks. However, existing endeavors predominately focus on the conventional semi-supervised setting where relatively abundant gold-labeled nodes are provided. While it is often impractical due to the fact that data labeling is unbearably laborious and requires intensive domain knowledge, especially when considering the heterogeneity of graph-structured data. Under the few-shot semi-supervised setting, the performance of most of the existing GNNs is inevitably undermined by the overfitting and oversmoothing issues, largely owing to the shortage of labeled data. In this paper, we propose a decoupled network architecture equipped with a novel meta-learning algorithm to solve this problem. In essence, our framework Meta-PN infers high-quality pseudo labels on unlabeled nodes via a meta-learned label propagation strategy, which effectively augments the scarce labeled data while enabling large receptive fields during training. Extensive experiments demonstrate that our approach offers easy and substantial performance gains compared to existing techniques on various benchmark datasets.

poster session 1 @ blue 1, poster session 8 @ blue 1, oral session 8 @ blue 1, poster session 1, poster session 8, oral session 8

It is well known that natural language models are vulnerable to adversarial attacks, which are mostly input-specific in nature. Recently, it has been shown that there also exist input-agnostic attacks in NLP models, called universal adversarial triggers. However, existing methods to craft universal triggersare data intensive. They require large amounts of data samplesto generate adversarial triggers, which are typically inaccessible by attackers. For instance, previous works take 3000 data samplesper class for the SNLI dataset to generate adversarial triggers. Inthis paper, we present a novel data-free approach, MINIMAL, tomine input-agnostic adversarial triggers from models. Using the triggers produced with our data-free algorithm, we reduce theaccuracy of Stanford Sentiment Treebank’s positive class from 93.6% to 9.6%. Similarly, for the Stanford Natural LanguageInference (SNLI), our single-word trigger reduces the accuracyof the entailment class from 90.95% to less than 0.6%. Despitebeing completely data-free, we get equivalent accuracy drops as data-dependent methods

poster session 1 @ red 4, poster session 8 @ red 4, poster session 1, poster session 8

We address cross-species 3D face morphing (i.e., 3D face morphing from human to animal), a novel problem with promising applications in social media and movie industry. It remains challenging how to preserve target structural information and source fine-grained facial details simultaneously. To this end, we propose an Alignment-aware 3D Face Morphing (AFM) framework, which builds semantic-adaptive correspondence between source and target faces across species, via an alignment-aware controller mesh (Explicit Controller, EC) with explicit source/target mesh binding. Based on EC, we introduce Controller-Based Mapping (CBM), which builds semantic consistency between source and target faces according to the semantic importance of different face regions. Additionally, an inference-stage coarse-to-fine strategy is exploited to produce fine-grained meshes with rich facial details from rough meshes. Extensive experimental results in multiple people and animals demonstrate that our method produces high-quality deformation results.

poster session 3 @ red 2, poster session 8 @ red 2, poster session 3, poster session 8

The development of InSAR (satellite Interferometric Synthetic Aperture Radar) enables accurate monitoring of land surface deformations, and has led to advances of deformation forecast for preventing landslide, which is one of the severe geological disasters. Despite the unparalleled success, existing spatio-temporal models typically make predictions on static adjacency relationships, simplifying the conditional dependencies and neglecting the distributions of variables. To overcome those limitations, we propose a Distribution Aware Probabilistic Framework (DAPF), which learns manifold embeddings while maintaining the distribution of deformations. We obtain a dynamic adjacency matrix upon which we approximate the true posterior while emphasizing the spatio-temporal characteristics. Experimental results on real-world dataset validate the superior performance of our method.

poster session 3 @ blue 2, poster session 8 @ blue 2, poster session 3, poster session 8

We present a novel approach to hierarchical reinforcement learning for linearly-solvable Markov decision processes. Our approach assumes that the state space is partitioned, and defines subtasks for moving between the partitions. We represent value functions on several levels of abstraction, and use the compositionality of subtasks to estimate the optimal values of the states in each partition. The policy is implicitly defined on these optimal value estimates, rather than being decomposed among the subtasks. As a consequence, our approach can learn the globally optimal policy, and does not suffer from non-stationarities induced by high-level decisions. If several partitions have equivalent dynamics, the subtasks of those partitions can be shared. We show that our approach is significantly more sample efficient than that of a flat learner and similar hierarchical approaches when the set of boundary states is smaller than the entire state space.

poster session 1 @ blue 1, poster session 8 @ blue 1, poster session 1, poster session 8

To deal with ambiguities in partial multi-label learning (PML), existing popular PML research attempts to perform disambiguation by direct ground-truth label identification. However, these approaches can be easily misled by noisy false-positive labels in the iteration of updating the model parameter and the latent ground-truth label variables. When labeling information is ambiguous, we should depend more on underlying structure of data, such as label and feature correlations, to perform disambiguation for partially labeled data. Moreover, large margin nearest neighbour (LMNN) is a popular strategy that considers data structure in classification. However, due to the ambiguity of labeling information in PML, traditional LMNN cannot be used to solve the PML problem directly. In addition, embedding is an effective technology to decrease the noise information of data. Inspried by LMNN and embedding technology, we propose a novel PML paradigm called Partial Multi-label Learning via Large Margin Nearest Neighbour Embeddings (PML-LMNNE), which aims to conduct disambiguation by projecting labels and features into a lower-dimension embedding space and reorganize the underlying structure by LMNN in the embedding space simultaneously. An efficient algorithm is designed to implement the proposed method and the convergence rate of the algorithm is analyzed. Moreover, we present a theoretical analysis of the generalization error bound for the proposed PML-LMNNE, which shows that the generalization error converges to the sum of two times the Bayes error over the labels when the number of instances $n$ goes to infinity. Comprehensive experiments on artificial and real-world datasets demonstrate the superiorities of the proposed PML-LMNNE.

poster session 3 @ blue 2, poster session 8 @ blue 2, oral session 3 @ blue 2, poster session 3, poster session 8, oral session 3

The cost-efficiency of model inference is critical to real-world machine learning (ML) applications, especially for delay-sensitive tasks and resource-limited devices. A typical dilemma is: in order to provide complex intelligent services (e.g. smart city), we need inference results of multiple ML models, but the cost budget (e.g. GPU memory) is not enough to run all of them. In this work, we dug into underlying relationships among black-box ML models and proposed a novel learning task: model linking. Model linking aims to bridges the knowledge of different black-box models by learning mappings between their output spaces. Based on model links, we developed a scheduling algorithm, named MLink. Through collaborative multi-model inference enabled by model links, MLink can significantly improve the accuracy of obtained inference results under the cost budget. We conducted comprehensive evaluations on a multi-modal dataset with seven different ML models and two real-world video analytics systems with 6 ML models and 3,264 hours of video. The experimental results show that our proposed model links can be effectively built among various black-box models. Under the budget of GPU memory, MLink can save 66.7% inference computations while preserving 94% inference accuracy, which outperforms multi-task learning, deep reinforcement learning-based scheduler and frame filtering baselines.

poster session 4 @ red 3, poster session 8 @ red 3, oral session 4 @ red 3, poster session 4, poster session 8, oral session 4

We study reward design strategies for incentivizing a reinforcement learning agent to adopt a policy from a set of admissible policies. The goal of the reward designer is to modify the underlying reward function cost-efficiently while ensuring that any approximately optimal deterministic policy under the new reward function is admissible and performs well under the original reward function. This problem can be viewed as a dual to the problem of optimal reward poisoning attacks: instead of forcing an agent to adopt a specific policy, the reward designer incentivizes an agent to avoid taking actions that are inadmissible in certain states. Perhaps surprisingly, and in contrast to the problem of optimal reward poisoning attacks, we first show that the reward design problem for admissible policy teaching is computationally challenging, and it is NP-hard to find an approximately optimal reward modification. We then proceed by formulating a surrogate problem whose optimal solution approximates the optimal solution to the reward design problem in our setting, but is more amenable to optimization techniques and analysis. For this surrogate problem, we present characterization results that provide bounds on the value of the optimal solution. Finally, we design a local search algorithm to solve the surrogate problem and showcase its utility using simulation-based experiments.

poster session 1 @ blue 1, poster session 8 @ blue 1, oral session 1 @ blue 1, poster session 1, poster session 8, oral session 1

Intent identification is the key technology in dialogue system. However, not all online queries are clear or complete. To identify users' intents from those fuzzy queries accurately, this paper proposes a multi-factor classification framework on the query level. Experimental results on our online serving system JIMI demonstrate the effectiveness of our proposed framework.

poster session 3 @ red 5, poster session 8 @ red 5, poster session 3, poster session 8

Graph-structured datasets usually have irregular graph sizes and connectivities, rendering the use of recent data augmentation techniques, such as Mixup, difficult. To tackle this challenge, we present the first Mixup-like graph augmentation method called Graph Transplant, which mixes irregular graphs in data space. To be well defined on various scales of the graph, our method identifies the sub-structure as a mix unit that can preserve the local information. Since the mixup-based methods without special consideration of the context are prone to generate noisy samples, our method explicitly employs the node saliency information to select meaningful subgraphs and adaptively determine the labels. We extensively validate our method with diverse GNN architectures on multiple graph classification benchmark datasets from a wide range of graph domains of different sizes. Experimental results show the consistent superiority of our method over other basic data augmentation baselines. We also demonstrate that Graph Transplant enhances the performance in terms of robustness and model calibration.

poster session 3 @ blue 2, poster session 8 @ blue 2, poster session 3, poster session 8

Unpaired video-to-video translation aims to translate videos between a source and a target domain without the need of paired training data, making it more feasible for real applications. Unfortunately, the translated videos generally suffer from temporal and semantic inconsistency. To address this, many existing works adopt spatiotemporal consistency constraints incorporating temporal information based on motion estimation. However, the inaccuracies in the estimation of motion deteriorate the quality of the guidance towards spatiotemporal consistency, which leads to unstable translation. In this work, we propose a novel paradigm that regularizes the spatiotemporal consistency by synthesizing motions in input videos with the generated optical flow instead of estimating them. Therefore, the synthetic motion can be applied in the regularization paradigm to keep motions consistent across domains without the risk of errors in motion estimation. Thereafter, we utilize our unsupervised recycle and unsupervised spatial loss, guided by the pseudo-supervision provided by the synthetic optical flow, to accurately enforce spatiotemporal consistency in both domains. Experiments show that our method is versatile in various scenarios and achieves state-of-the-art performance in generating temporally and semantically consistent videos. Code is available at: https://github.com/wangkaihong/Unsup_Recycle_GAN/.

poster session 3 @ red 1, poster session 8 @ red 1, poster session 3, poster session 8

Reinforcement Learning (RL) agents in the real world must satisfy safety constraints in addition to maximizing a reward objective. Model-based RL algorithms hold promise for reducing unsafe real-world actions: they may synthesize policies that obey all constraints using simulated samples from a learned model. However, imperfect models can result in real-world constraint violations even for actions that are predicted to satisfy all constraints. We propose Conservative and Adaptive Penalty (CAP), a model-based safe RL framework that accounts for potential modeling errors by capturing model uncertainty and adaptively exploiting it to balance the reward and the cost objectives. First, CAP inflates predicted costs using an uncertainty-based penalty. Theoretically, we show that policies that satisfy this conservative cost constraint are guaranteed to also be feasible in the true environment. We further show that this guarantees the safety of all intermediate solutions during RL training. Further, CAP adaptively tunes this penalty during training using true cost feedback from the environment. We evaluate this conservative and adaptive penalty-based approach for model-based safe RL extensively on state and image-based environments. Our results demonstrate substantial gains in sample-efficiency while incurring fewer violations than prior safe RL algorithms. Code is available at: https://github.com/Redrew/CAP

poster session 1 @ blue 1, poster session 8 @ blue 1, poster session 1, poster session 8