IJCAI 2023 Papers

Incorporating physics in human motion capture to avoid artifacts like floating, foot sliding, and ground penetration is a promising direction. Existing solutions always adopt kinematic results as reference motions, and the physics is treated as a post-processing module. However, due to the depth ambiguity, monocular motion capture inevitably suffers from noises, and the noisy reference often leads to failure for physics-based tracking. To address the obstacles, our key-idea is to employ physics as denoising guidance in the reverse diffusion process to reconstruct physically plausible human motion from a modeled pose probability distribution. Specifically, we first train a latent gaussian model that encodes the uncertainty of 2D-to-3D lifting to facilitate reverse diffusion. Then, a physics module is constructed to track the motion sampled from the distribution. The discrepancies between the tracked motion and image observation are used to provide explicit guidance for the reverse diffusion model to refine the motion. With several iterations, the physics-based tracking and kinematic denoising promote each other to generate a physically plausible human motion. Experimental results show that our method outperforms previous physics-based methods in both joint accuracy and success rate. More information can be found at https://github.com/Me-Ditto/Physics-Guided-Mocap.

The subgraph isomorphism problem (SIP) is a challenging problem with wide practical applications. In the last decade, despite being a theoretical hard problem, researchers design various algorithms for solving SIP. In this work, we propose three main heuristics and develop an improved exact algorithm for SIP. First, we design a probing search procedure to try whether the search procedure can successfully obtain a solution at first sight. Second, we design a novel matching ordering as a value-ordering heuristic, which uses some useful information obtained from the probing search procedure to preferentially select some promising target vertices. Third, we discuss the characteristics of different propagation methods in the context of SIP and present an adaptive propagation method to make a good balance between these methods. Experimental results on a broad range of real-world benchmarks show that our proposed algorithm performs better than state-of-the-art algorithms for the SIP.

Recent work for extracting relations from texts has achieved excellent performance. However, existing studies mainly focus on simple relation extraction, these methods perform not well on overlapping triple problem because the tags of shared entities would conflict with each other. Especially, overlapping entities are common and indispensable in Chinese. To address this issue, this paper proposes PasCore, which utilizes a global pointer annotation strategy for overlapping relation extraction in Chinese. PasCore first obtains the sentence vector via general pre-training model encoder, and uses classifier to predicate relations. Subsequently, it uses global pointer annotation strategy for head entity annotation, which uses global tags to label the start and end positions of the entities. Finally, PasCore integrates the relation, head entity and its type to mark the tail entity. Furthermore, PasCore performs conditional layer normalization to fuse features, which connects all stages and greatly enriches the association between relations and entities. Experimental results on both Chinese and English real-world datasets demonstrate that PasCore outperforms strong baselines on relation extraction and, especially, shows superior performance on overlapping relation extraction.

We provide a library of participatory budgeting data (Pabulib) and open source tools (Pabutools and Pabustats) for analysing this data. We analyse how the results of participatory budgeting elections would change if a different selection rule was applied. We provide evidence that the outcomes of the Method of Equal Shares would be considerably fairer than those of the Utilitarian Greedy rule that is currently in use. We also show that the division of the projects into districts and/or categories can in many cases be avoided when using proportional rules. We find that this would increase the overall utility of the voters.

In an indivisible participatory budgeting (PB) framework, we have a limited budget that is to be distributed among a set of projects, by aggregating the preferences of voters for the projects. All the prior work on indivisible PB assumes that each project has only one possible cost. In this work, we let each project have a set of permissible costs, each reflecting a possible degree of sophistication of the project. Each voter approves a range of costs for each project, by giving an upper and lower bound on the cost that she thinks the project deserves. The outcome of a PB rule selects a subset of projects and also specifies their corresponding costs. We study different utility notions and prove that the existing positive results when every project has exactly one permissible cost can also be extended to our framework where a project has several permissible costs. We also analyze the fixed parameter tractability of the problem. Finally, we propose some important and intuitive axioms and analyze their satisfiability by different PB rules. We conclude by making some crucial remarks.

In recent years, remarkable results have been achieved in self-supervised action recognition using skeleton sequences with contrastive learning. It has been observed that the semantic distinction of human action features is often represented by local body parts, such as legs or hands, which are advantageous for skeleton-based action recognition. This paper proposes an attention-based contrastive learning framework for skeleton representation learning, called SkeAttnCLR, which integrates local similarity and global features for skeleton-based action representations. To achieve this, a multi-head attention mask module is employed to learn the soft attention mask features from the skeletons, suppressing non-salient local features while accentuating local salient features, thereby bringing similar local features closer in the feature space. Additionally, ample contrastive pairs are generated by expanding contrastive pairs based on salient and non-salient features with global features, which guide the network to learn the semantic representations of the entire skeleton. Therefore, with the attention mask mechanism, SkeAttnCLR learns local features under different data augmentation views. The experiment results demonstrate that the inclusion of local feature similarity significantly enhances skeleton-based action representation. Our proposed SkeAttnCLR outperforms state-of-the-art methods on NTURGB+D, NTU120-RGB+D, and PKU-MMD datasets. The code and settings are available at this repository: https://github.com/GitHubOfHyl97/SkeAttnCLR.

Compensation is a strategy that a semantics may follow when it faces dilemmas between quality and quantity of attackers. It allows several weak attacks to compensate one strong attack. It is based on compensation degree, which is a tuple that indicates (i) to what extent an attack is weak and (ii) the number of weak attacks needed to compensate a strong one. Existing principles on compensation do not specify the parameters, thus it is unclear whether semantics satisfying them compensate at only one degree or several degrees, and which ones. This paper proposes a parameterised family of gradual semantics, which unifies multiple semantics that share some principles but differ in their strategy regarding solving dilemmas. Indeed, we show that the two semantics taking the extreme values of the parameter favour respectively quantity and quality, while all the remaining ones compensate at some degree. We define three classes of compensation degrees and show that the novel family is able to compensate at all of them while none of the existing gradual semantics does.

In the NP-hard Max c-Cut problem, one is given an undirected edge-weighted graph G and wants to color the vertices of G with c colors such that the total weight of edges with distinctly colored endpoints is maximal. The case with c=2 is the famous Max Cut problem. To deal with the NP-hardness of this problem, we study parameterized local search algorithms. More precisely, we study LS-Max c-Cut where we are additionally given a vertex coloring f and an integer k and the task is to find a better coloring f' that differs from f in at most k entries, if such a coloring exists; otherwise, f is k-optimal. We show that LS-Max c-Cut presumably cannot be solved in g(k) · nᴼ⁽¹⁾ time even on bipartite graphs, for all c ≥ 2. We then show an algorithm for LS-Max c-Cut with running time O((3eΔ)ᵏ · c · k³ · Δ · n), where Δ is the maximum degree of the input graph. Finally, we evaluate the practical performance of this algorithm in a hill-climbing approach as a post-processing for state-of-the-art heuristics for Max c-Cut. We show that using parameterized local search, the results of this heuristic can be further improved on a set of standard benchmark instances.

Event Causality Identification (ECI) aims to identify the causality between a pair of event mentions in a document, which is composed of sentence-level ECI (SECI) and document-level ECI (DECI). Previous work applies various reasoning models to identify the implicit event causality. However, they indiscriminately reason all event causality in the same way, ignoring that most inter-sentence event causality depends on intra-sentence event causality to infer. In this paper, we propose a progressive graph pairwise attention network (PPAT) to consider the above dependence. PPAT applies a progressive reasoning strategy, as it first predicts the intra-sentence event causality, and then infers the more implicit inter-sentence event causality based on the SECI result. We construct a sentence boundary event relational graph, and PPAT leverages a simple pairwise attention mechanism, which attends to different reasoning chains on the graph. In addition, we propose a causality-guided training strategy for assisting PPAT in learning causality-related representations on every layer. Extensive experiments show that our model achieves state-of-the-art performance on three benchmark datasets (5.5%, 2.2% and 4.5% F1 gains on EventStoryLine, MAVEN-ERE and Causal-TimeBank). Code is available at https://github.com/HITsz-TMG/PPAT.

The recent rise in popularity of Hyperparameter Optimization (HPO) for deep learning has highlighted the role that good hyperparameter (HP) space design can play in training strong models. In turn, designing a good HP space is critically dependent on understanding the role of different HPs. This motivates research on HP Importance (HPI), e.g., with the popular method of functional ANOVA (f-ANOVA). However, the original f-ANOVA formulation is inapplicable to the subspaces most relevant to algorithm designers, such as those defined by top performance. To overcome this issue, we derive a novel formulation of f-ANOVA for arbitrary subspaces and propose an algorithm that uses Pearson divergence (PED) to enable a closed-form calculation of HPI. We demonstrate that this new algorithm, dubbed PED-ANOVA, is able to successfully identify important HPs in different subspaces while also being extremely computationally efficient. See https://arxiv.org/abs/2304.10255 for the latest version with Appendix.

The World Health Organization has underlined the significance of expediting the preventive measures for crime against women and children to attain the United Nations Sustainable Development Goals 2030 (promoting well-being, gender equality, and equal access to justice). The crime victims typically need mental health and legal counselling support for their ultimate well-being and sometimes they need to be persuaded to seek desired support. Further, counselling interactions should adopt correct politeness and empathy strategies so that a warm, amicable, and respectful environment can be built to better understand the victims’ situations. To this end, we propose PARTNER, a Politeness and empAthy strategies-adaptive peRsuasive dialogue sysTem for meNtal health and LEgal counselling of cRime victims. For this, first, we create a novel mental HEalth and legAl counseLling conversational dataset HEAL, annotated with three distinct aspects, viz. counselling act, politeness strategy, and empathy strategy. Then, by formulating a novel reward function, we train a counselling dialogue system in a reinforcement learning setting to ensure correct counselling act, politeness strategy, and empathy strategy in the generated responses. Extensive empirical analysis and experimental results show that the proposed reward function ensures persuasive counselling responses with correct polite and empathetic tone in the generated responses. Further, PARTNER proves its efficacy to engage the victim by generating diverse and natural responses.

Learning disentangled representations with variational autoencoders (VAEs) is often attributed to the regularisation component of the loss. In this work, we highlight the interaction between data and the reconstruction term of the loss as the main contributor to disentanglement in VAEs. We show that standard benchmark datasets have unintended correlations between their subjective ground-truth factors and perceived axes in the data according to typical VAE reconstruction losses. Our work exploits this relationship to provide a theory for what constitutes an adversarial dataset under a given reconstruction loss. We verify this by constructing an example dataset that prevents disentanglement in state-of-the-art frameworks while maintaining human-intuitive ground-truth factors. Finally, we re-enable disentanglement by designing an example reconstruction loss that is once again able to perceive the ground-truth factors. Our findings demonstrate the subjective nature of disentanglement and the importance of considering the interaction between the ground-truth factors, data and notably, the reconstruction loss, which is under-recognised in the literature.

We study a scenario where an adjudication task (e.g., the resolution of a binary dispute) is outsourced to a set of agents who are appointed as jurors. This scenario is particularly relevant in a Web3 environment, where no verification of the adjudication outcome is possible, and the appointed agents are, in principle, indifferent to the final verdict. We consider simple adjudication mechanisms that use (1) majority voting to decide the final verdict and (2) a payment function to reward the agents with the majority vote and possibly punish the ones in the minority. Agents interact with such a mechanism strategically: they exert some effort to understand how to properly judge the dispute and cast a yes/no vote that depends on this understanding and on information they have about the rest of the votes. Eventually, they vote so that their utility (i.e., their payment from the mechanism minus the cost due to their effort) is maximized. Under reasonable assumptions about how an agent's effort is related to her understanding of the dispute, we show that appropriate payment functions can be used to recover the correct adjudication outcome with high probability. Our findings follow from a detailed analysis of the induced strategic game and make use of both theoretical arguments and simulation experiments.

Widely-used DNN models are vulnerable to backdoor attacks, where the backdoored model is only triggered by specific inputs but can maintain a high prediction accuracy on benign samples. Existing backdoor input detection strategies rely on the assumption that benign and poisoned samples are separable in the feature representation of the model. However, such an assumption can be broken by advanced feature-hidden backdoor attacks. In this paper, we propose a novel detection framework, dubbed Orion (online backdoor sample detection via evolution deviance). Specifically, we analyze how predictions evolve during a forward pass and find deviations between the shallow and deep outputs of the backdoor inputs. By introducing side nets to track such evolution divergence, Orion eliminates the need for the assumption of latent separability. Additionally, we put forward a scheme to restore the original label of backdoor samples, enabling more robust predictions. Extensive experiments on six attacks, three datasets, and two architectures verify the effectiveness of Orion. It is shown that Orion outperforms state-of-the-art defenses and can identify feature-hidden attacks with an F1-score of 90%, compared to 40% for other detection schemes. Orion can also achieve 80% label recovery accuracy on basic backdoor attacks.

Scene text recognition (STR) has attracted much attention due to its broad applications. The previous works pay more attention to dealing with the recognition of Latin text images with complex backgrounds by introducing language models or other auxiliary networks. Different from Latin texts, many vertical Chinese texts exist in natural scenes, which brings difficulties to current state-of-the-art STR methods. In this paper, we take the first attempt to extract orientation-independent visual features by disentangling content and orientation information of text images, thus recognizing both horizontal and vertical texts robustly in natural scenes. Specifically, we introduce a Character Image Reconstruction Network (CIRN) to recover corresponding printed character images with disentangled content and orientation information. We conduct experiments on a scene dataset for benchmarking Chinese text recognition, and the results demonstrate that the proposed method can indeed improve performance through disentangling content and orientation information. To further validate the effectiveness of our method, we additionally collect a Vertical Chinese Text Recognition (VCTR) dataset. The experimental results show that the proposed method achieves 45.63\% improvement on VCTR when introducing CIRN to the baseline model.

In fair division of indivisible goods, l-out-of-d maximin share (MMS) is the value that an agent can guarantee by partitioning the goods into d bundles and choosing the l least preferred bundles. Most existing works aim to guarantee to all agents a constant fraction of their 1-out-of-n MMS. But this guarantee is sensitive to small perturbation in agents' cardinal valuations. We consider a more robust approximation notion, which depends only on the agents' ordinal rankings of bundles. We prove the existence of l-out-of-floor((l+1/2)n) MMS allocations of goods for any integer l greater than or equal to 1, and present a polynomial-time algorithm that finds a 1-out-of-ceiling(3n/2) MMS allocation when l = 1. We further develop an algorithm that provides a weaker ordinal approximation to MMS for any l > 1.

We study a variant of hedonic games, called hedonic seat arrangements in the literature, where the goal is not to partition the agents into coalitions but to assign them to vertices of a given graph; their satisfaction is then based on the subset of agents in their neighborhood. We focus on ordinal hedonic seat arrangements where the preferences over neighborhoods are deduced from ordinal preferences over single agents and a given preference extension. In such games and for different types of preference restrictions and extensions, we investigate the existence of arrangements satisfying stability w.r.t. swaps of positions in the graph or the well-known optimality concept of popularity.

One of the factors that hinder the adoption of nonmonotonic description logics in applications is performance. Even when monotonic and nonmonotonic inferences have the same asymptotic complexity, the implementation of nonmonotonic reasoning may be significantly slower. The family of nonmonotonic logics DLN is no exception to this behavior. We address this issue by introducing two provably correct and complete optimizations for reasoning in DLN. The first optimization is a module extractor that has the purpose of focusing reasoning on a relevant subset of the knowledge base. The second, called optimistic evaluation, aims at exploiting incremental reasoning in a better way. Extensive experimental evaluation shows that the optimized DLN reasoning is often compatible with interactive query answering, thus bringing nonmonotonic description logics closer to practical applications.

Crop management has a significant impact on crop yield, economic profit, and the environment. Although management guidelines exist, finding the optimal management practices is challenging. Previous work used reinforcement learning (RL) and crop simulators to solve the problem, but the trained policies either have limited performance or are not deployable in the real world. In this paper, we present an intelligent crop management system that optimizes nitrogen fertilization and irrigation simultaneously via RL, imitation learning (IL), and crop simulations using the Decision Support System for Agrotechnology Transfer (DSSAT). We first use deep RL, in particular, deep Q-network, to train management policies that require a large number of state variables from the simulator as observations (denoted as full observation). We then invoke IL to train management policies that only need a few state variables that can be easily obtained or measured in the real world (denoted as partial observation) by mimicking the actions of the RL policies trained under full observation. Simulation experiments using the maize crop in Florida (US) and Zaragoza (Spain) demonstrate that the trained policies from both RL and IL techniques achieved more than 45\% improvement in economic profit while causing less environmental impact compared with a baseline method. Most importantly, the IL-trained management policies are directly deployable in the real world as they use readily available information.