IJCAI 2022 论文列表

Efficient algorithms computing a Nash equilibrium have been successfully applied to large zero- sum two-player extensive-form games (e.g., poker). However, in multiplayer games, computing a Nash equilibrium is generally hard, and the equilibria are not exchangeable, which makes players face the problem of selecting one of many different Nash equilibria. In this paper, we focus on an alternative solution concept in zero-sum multiplayer extensive-form games called Team-Maxmin Equilibrium (TME). It is a Nash equilibrium that maximizes each team member’s utility. As TME is unique in general, it avoids the equilibrium selection problem. However, it is still difficult (FNP- hard) to find a TME. Computing it can be formulated as a non-convex program, but existing algorithms are capable of solving this program for only very small games. In this paper, we first refine the complexity result for computing a TME by using a correlation plan to show that a TME can be found in polynomial time in a specific class of games according to our boundary for complexity. Second, we propose an efficient correlation-based algorithm to solve the non-convex program for TME in games not belonging to this class. The algorithm combines two special correlation plans based on McCormick envelopes for convex relaxation and von Stengel-Forges polytope for correlated equilibria. We show that restricting the feasible solution space to von Stengel-Forges polytope will strictly reduce the feasible solution space after convex re- laxation of nonlinear terms. Finally, experiments show that our algorithm is about four orders of magnitude faster than the prior state of the art and can solve many previously unsolvable games.

Corner-guided detector enjoys potential ability to yield precise bounding boxes. However, unreliable corner pairs, generated by heuristic grouping guidance, hinder the development of this detector. In this paper, we propose a novel corner grouping algorithm, termed as Corner Affinity, to significantly boost the reliability and robustness of corner grouping. The proposed Corner Affinity is a couple of two interactional factors, namely, 1) the structure affinity (SA), applying to generate preliminary corner pairs through the corresponding object's shallow construction information. 2) the contexts affinity (CA), running as optimizing corner pairs via embedding deeper semantic features of affiliated instances. Equipped with the Corner Affinity, a detector can produce high-quality bounding boxes upon preferable paired corner keypoints. Experimental results show the superiority of our design on multiple benchmark datasets. Specifically, for CornerNet baseline, the proposed Corner Affinity brings AP boostings of 5.8% on COCO, 35.8% on Citypersons, and 17.2% on UCAS-AOD without bells and whistles.

One compelling application of artificial intelligence is to generate a video of a target person performing arbitrary desired motion (from a source person). While the state-of-the-art methods are able to synthesize a video demonstrating similar broad stroke motion details, they are generally lacking in texture details. A pertinent manifestation appears as distorted face, feet, and hands, and such flaws are very sensitively perceived by human observers. Furthermore, current methods typically employ GANs with a L2 loss to assess the authenticity of the generated videos, inherently requiring a large amount of training samples to learn the texture details for adequate video generation. In this work, we tackle these challenges from three aspects: 1) We disentangle each video frame into foreground (the person) and background, focusing on generating the foreground to reduce the underlying dimension of the network output. 2) We propose a theoretically motivated Gromov-Wasserstein loss that facilitates learning the mapping from a pose to a foreground image. 3) To enhance texture details, we encode facial features with geometric guidance and employ local GANs to refine the face, feet, and hands. Extensive experiments show that our method is able to generate realistic target person videos, faithfully copying complex motions from a source person. Our code and datasets are released at https://github.com/Sifann/FakeMotion.

Conversational semantic role labeling (CSRL) is a newly proposed task that uncovers the shallow semantic structures in a dialogue text. Unfortunately several important characteristics of the CSRL task have been overlooked by the existing works, such as the structural information integration, near-neighbor influence. In this work, we investigate the integration of a latent graph for CSRL. We propose to automatically induce a predicate-oriented latent graph (POLar) with a predicate-centered gaussian mechanism, by which the nearer and informative words to the predicate will be allocated with more attention. The POLar structure is then dynamically pruned and refined so as to best fit the task need. We additionally introduce an effective dialogue-level pre-trained language model, CoDiaBERT, for better supporting multiple utterance sentences and handling the speaker coreference issue in CSRL. Our system outperforms best-performing baselines on three benchmark CSRL datasets with big margins, especially achieving over 4% F1 score improvements on the cross-utterance argument detection. Further analyses are presented to better understand the effectiveness of our proposed methods.

We present a novel dataset of dialogues containing interruption with an aim to conduct a large-scale analysis of interruption patterns of people from diverse backgrounds in terms of gender, race/ethnicity, occupation, and political orientation. Our dataset includes 625,409 dialogues containing interruptions found in 275,420 transcripts from CNN, Fox News, and MSNBC spanning between January 2000 and July 2021. From this large, unlabeled pool of interruptions, we release an annotated dataset consisting of 2,000 dialogues with fine-grained interruption labels. We use this dataset to train an interruption classifier and predict the interruption type of a given dialogue. Our results reveal that male speakers (in our collected samples) tend to talk more than female speakers, while female speakers interrupt more. Moreover, people tend to use less intrusive interruptions when talking to others sharing the same political belief. This pattern becomes more pronounced among news media with stronger political bias.

Recent advances in large pre-trained language models have demonstrated strong results in generating natural languages and significantly improved performances for many natural language generation (NLG) applications such as machine translation and text summarization. However, when the generation tasks are more open-ended and the content is under-specified, existing techniques struggle to generate long-term coherent and creative content. Moreover, the models exhibit and even amplify social biases that are learned from the training corpora. This happens because the generation models are trained to capture the surface patterns (i.e. sequences of words), instead of capturing underlying semantics and discourse structures, as well as background knowledge including social norms. In this paper, I introduce our recent works on controllable text generation to enhance the creativity and fairness of language generation models. We explore hierarchical generation and constrained decoding, with applications to creative language generation including story, poetry, and figurative languages, and bias mitigation for generation models.

Emotional support conversation aims at reducing the emotional distress of the help-seeker, which is a new and challenging task. It requires the system to explore the cause of help-seeker's emotional distress and understand their psychological intention to provide supportive responses. However, existing methods mainly focus on the sequential contextual information, ignoring the hierarchical relationships with the global cause and local psychological intention behind conversations, thus leads to a weak ability of emotional support. In this paper, we propose a Global-to-Local Hierarchical Graph Network to capture the multi-source information (global cause, local intentions and dialog history) and model hierarchical relationships between them, which consists of a multi-source encoder, a hierarchical graph reasoner, and a global-guide decoder. Furthermore, a novel training objective is designed to monitor semantic information of the global cause. Experimental results on the emotional support conversation dataset, ESConv, confirm that the proposed GLHG has achieved the state-of-the-art performance on the automatic and human evaluations.

Hierarchical clustering recursively partitions data at an increasingly finer granularity. In real-world applications, multi-view data have become increasingly important. This raises a less investigated problem, i.e., multi-view hierarchical clustering, to better understand the hierarchical structure of multi-view data. To this end, we propose a novel neural network-based model, namely Contrastive Multi-view Hyperbolic Hierarchical Clustering(CMHHC). It consists of three components, i.e., multi-view alignment learning, aligned feature similarity learning, and continuous hyperbolic hierarchical clustering. First, we align sample-level representations across multiple views in a contrastive way to capture the view-invariance information. Next, we utilize both the manifold and Euclidean similarities to improve the metric property. Then, we embed the representations into a hyperbolic space and optimize the hyperbolic embeddings via a continuous relaxation of hierarchical clustering loss. Finally, a binary clustering tree is decoded from optimized hyperbolic embeddings. Experimental results on five real-world datasets demonstrate the effectiveness of the proposed method and its components.

Personality detection is to identify the personality traits underlying social media posts. Most of the existing work is mainly devoted to learning the representations of posts based on labeled data. Yet the ground-truth personality traits are collected through time-consuming questionnaires. Thus, one of the biggest limitations lies in the lack of training data for this data-hungry task. In addition, the correlations among traits should be considered since they are important psychological cues that could help collectively identify the traits. In this paper, we construct a fully-connected post graph for each user and develop a novel Contrastive Graph Transformer Network model (CGTN) which distills potential labels of the graphs based on both labeled and unlabeled data. Specifically, our model first explores a self-supervised Graph Neural Network (GNN) to learn the post embeddings. We design two types of post graph augmentations to incorporate different priors based on psycholinguistic knowledge of Linguistic Inquiry and Word Count (LIWC) and post semantics. Then, upon the post embeddings of the graph, a Transformer-based decoder equipped with post-to-trait attention is exploited to generate traits sequentially. Experiments on two standard datasets demonstrate that our CGTN outperforms the state-of-the-art methods for personality detection.

Despite the remarkable progress of deep learning models for semantic segmentation, the success of these models is strongly limited by the following aspects: 1) large datasets with pixel-level annotations must be available and 2) training must be performed with all classes simultaneously. Indeed, in incremental learning scenarios, where new classes are added to an existing framework, these models are prone to catastrophic forgetting of previous classes. To address these two limitations, we propose a weakly-supervised mechanism for continual semantic segmentation that can leverage cheap image-level annotations and a novel rehearsal strategy that intertwines the learning of past and new classes. Specifically, we explore two rehearsal technique variants: 1) imprinting past objects on new images and 2) transferring past representations in intermediate features maps. We conduct extensive experiments on Pascal-VOC by varying the proportion of fully- and weakly-supervised data in various setups and show that our contributions consistently improve the mIoU on both past and novel classes. Interestingly, we also observe that models trained with less data in incremental steps sometimes outperform the same architectures trained with more data. We discuss the significance of these results and propose some hypotheses regarding the dynamics between forgetting and learning.

Federated learning (FL) is a promising approach for learning a shared global model on decentralized data owned by multiple clients without exposing their privacy. In real-world scenarios, data accumulated at the client-side varies in distribution over time. As a consequence, the global model tends to forget the knowledge obtained from previous tasks while learning new tasks, showing signs of "catastrophic forgetting". Previous studies in centralized learning use techniques such as data replay and parameter regularization to mitigate catastrophic forgetting. Unfortunately, these techniques cannot adequately solve the non-trivial problem in FL. We propose Continual Federated Learning with Distillation (CFeD) to address catastrophic forgetting under FL. CFeD performs knowledge distillation on both the clients and the server, with each party independently having an unlabeled surrogate dataset, to mitigate forgetting. Moreover, CFeD assigns different learning objectives, namely learning the new task and reviewing old tasks, to different clients, aiming to improve the learning ability of the model. The results show that our method performs well in mitigating catastrophic forgetting and achieves a good trade-off between the two objectives.

We study how to incentivize agents in a target subpopulation to produce a higher output by means of rank-order allocation contests, in the context of incomplete information. We describe a symmetric Bayes--Nash equilibrium for contests that have two types of rank-based prizes: (1) prizes that are accessible only to the agents in the target group; (2) prizes that are accessible to everyone. We also specialize this equilibrium characterization to two important sub-cases: (i) contests that do not discriminate while awarding the prizes, i.e., only have prizes that are accessible to everyone; (ii) contests that have prize quotas for the groups, and each group can compete only for prizes in their share. For these models, we also study the properties of the contest that maximizes the expected total output by the agents in the target group.

Anomaly detection aims to separate anomalies from normal samples, and the pretrained network is promising for anomaly detection. However, adapting the pretrained features would be confronted with the risk of pattern collapse when finetuning on one-class training data. In this paper, we propose an anomaly detection framework called constrained adaptive projection with pretrained features (CAP). Combined with pretrained features, a simple linear projection head applied on a specific input and its k most similar pretrained normal representations is designed for feature adaptation, and a reformed self-attention is leveraged to mine the inner-relationship among one-class semantic features. A loss function is proposed to avoid potential pattern collapse. Concretely, it considers the similarity between a specific data and its corresponding adaptive normal representation, and incorporates a constraint term slightly aligning pretrained and adaptive spaces. Our method achieves state-of-the-art anomaly detection performance on semantic anomaly detection and sensory anomaly detection benchmarks including 96.5% AUROC on CIFAR-100 dataset, 97.0% AUROC on CIFAR-10 dataset and 89.9% AUROC on MvTec dataset.

Should the properties of constraint monotonicity and foundedness be mandatory requirements that every answer set and world view semantics must satisfy? This question is challenging and has incurred a debate in answer set programming (ASP). In this paper we address the question by introducing natural logic programs whose expected answer sets and world views violate these properties and thus may be viewed as counter-examples to these requirements. Specifically we use instances of the generalized strategic companies problem for ASP benchmark competitions as concrete examples to demonstrate that the requirements of constraint monotonicity and foundedness may exclude expected answer sets for some simple disjunctive programs and world views for some epistemic specifications. In conclusion these properties should not be mandatory conditions for an answer set and world view semantics in general.

Conditional independence is a crucial concept for efficient probabilistic reasoning. For symbolic and qualitative reasoning, however, it has played only a minor role. Recently, Lynn, Delgrande, and Peppas have considered conditional independence in terms of syntactic multivalued dependencies. In this paper, we define conditional independence as a semantic property of epistemic states and present axioms for iterated belief revision operators to obey conditional independence in general. We show that c-revisions for ranking functions satisfy these axioms, and exploit the relevance of these results for iterated belief revision in general.

Although heuristic search is one of the most successful approaches to classical planning, this planning paradigm does not apply straightforwardly to Generalized Planning (GP). This paper adapts the planning as heuristic search paradigm to the particularities of GP, and presents the first native heuristic search approach to GP. First, the paper defines a program-based solution space for GP that is independent of the number of planning instances in a GP problem, and the size of these instances. Second, the paper defines the BFGP algorithm for GP, that implements a best-first search in our program-based solution space, and that is guided by different evaluation and heuristic functions.

Classical instance queries over an ontology only consider explicitly named individuals. Concept referring expressions (CREs) also allow for returning answers in the form of concepts that describe implicitly given individuals in terms of their relation to an explicitly named one. Existing approaches, e.g., based on tree automata, can neither be integrated into state-of-the-art OWL reasoners nor are they directly amenable for an efficient implementation. To address this, we devise a novel algorithm that uses highly optimized OWL reasoners as a black box. In addition to the standard criteria of singularity and certainty for CREs, we devise and consider the criterion of uniqueness of CREs for Horn ALC ontologies. The evaluation of our prototypical implementation shows that computing CREs for the most general concept (Top) can be done in less than one minute for ontologies with thousands of individuals and concepts.

Computing the gradient of stochastic Plackett-Luce (PL) ranking models for relevance and fairness metrics can be infeasible because it requires iterating over all possible permutations of items. In this paper, we introduce a novel algorithm: PL-Rank, that estimates the gradient of a PL ranking model through sampling. Unlike existing approaches, PL-Rank makes use of the specific structure of PL models and ranking metrics. Our experimental analysis shows that PL-Rank has a greater sample-efficiency and is computationally less costly than existing policy gradients, resulting in faster convergence at higher performance.

In this paper, we study the graphic layout generation problem of producing high-quality visual-textual presentation designs for given images. We note that image compositions, which contain not only global semantics but also spatial information, would largely affect layout results. Hence, we propose a deep generative model, dubbed as composition-aware graphic layout GAN (CGL-GAN), to synthesize layouts based on the global and spatial visual contents of input images. To obtain training images from images that already contain manually designed graphic layout data, previous work suggests masking design elements (e.g., texts and embellishments) as model inputs, which inevitably leaves hint of the ground truth. We study the misalignment between the training inputs (with hint masks) and test inputs (without masks), and design a novel domain alignment module (DAM) to narrow this gap. For training, we built a large-scale layout dataset which consists of 60,548 advertising posters with annotated layout information. To evaluate the generated layouts, we propose three novel metrics according to aesthetic intuitions. Through both quantitative and qualitative evaluations, we demonstrate that the proposed model can synthesize high-quality graphic layouts according to image compositions. The data and code will be available at https://github.com/minzhouGithub/CGL-GAN.