IJCAI 2023 Papers

Commonsense Question Answering (CQA) aims to answer questions that require human commonsense. Closed-book CQA, as one of the subtasks, requires the model to answer questions without retrieving external knowledge, which emphasizes the importance of the model's problem-solving ability. Most previous methods relied on large-scale pre-trained models to generate question-related knowledge while ignoring the crucial role of skills in the process of answering commonsense questions. Generally, skills refer to the learned ability in performing a specific task or activity, which are derived from knowledge and experience. In this paper, we introduce a new approach named Dynamic Skill-aware Commonsense Question Answering (DSCQA), which transcends the limitations of traditional methods by informing the model about the need for each skill in questions and utilizes skills as a critical driver in CQA process. To be specific, DSCQA first employs commonsense skill extraction module to generate various skill representations. Then, DSCQA utilizes dynamic skill module to generate dynamic skill representations. Finally, in perception and emphasis module, various skills and dynamic skill representations are used to help question-answering process. Experimental results on two publicly available CQA datasets show the effectiveness of our proposed model and the considerable impact of introducing skills.

Recently, Zero-Shot Node Classification (ZNC) has been an emerging and crucial task in graph data analysis. This task aims to predict nodes from unseen classes which are unobserved in the training process. Existing work mainly utilizes Graph Neural Networks (GNNs) to associate features' prototypes and labels' semantics thus enabling knowledge transfer from seen to unseen classes. However, the multi-faceted semantic orientation in the feature-semantic alignment has been neglected by previous work, i.e. the content of a node usually covers diverse topics that are relevant to the semantics of multiple labels. It's necessary to separate and judge the semantic factors that tremendously affect the cognitive ability to improve the generality of models. To this end, we propose a Knowledge-Aware Multi-Faceted framework (KMF) that enhances the richness of label semantics via the extracted KG (Knowledge Graph)-based topics. And then the content of each node is reconstructed to a topic-level representation that offers multi-faceted and fine-grained semantic relevancy to different labels. Due to the particularity of the graph's instance (i.e., node) representation, a novel geometric constraint is developed to alleviate the problem of prototype drift caused by node information aggregation. Finally, we conduct extensive experiments on several public graph datasets and design an application of zero-shot cross-domain recommendation. The quantitative results demonstrate both the effectiveness and generalization of KMF with the comparison of state-of-the-art baselines.

Self-training (ST) has come to fruition in language understanding tasks by producing pseudo labels, which reduces the labeling bottleneck of language model fine-tuning. Nevertheless, in facilitating semi-supervised controllable language generation, ST faces two key challenges. First, augmented by self-generated pseudo text, generation models tend to over-exploit the previously learned text distribution, suffering from mode collapse and poor generation diversity. Second, generating pseudo text in each iteration is time-consuming, severely decelerating the training process. In this work, we propose KEST, a novel and efficient self-training framework to handle these problems. KEST utilizes a kernel-based loss, rather than standard cross entropy, to learn from the soft pseudo text produced by a shared non-autoregressive generator. We demonstrate both theoretically and empirically that KEST can benefit from more diverse pseudo text in an efficient manner, which allows not only refining and exploiting the previously fitted distribution but also enhanced exploration towards a larger potential text space, providing a guarantee of improved performance. Experiments on three controllable generation tasks demonstrate that KEST significantly improves control accuracy while maintaining comparable text fluency and generation diversity against several strong baselines.

In recent years, graph Transformers (GTs) have been demonstrated as a robust architecture for a wide range of graph learning tasks. However, the quadratic complexity of GTs limits their scalability on large-scale data, in comparison to Graph Neural Networks (GNNs). In this work, we propose the Kernel Decomposition Linear Graph Transformer (KDLGT), an accelerating framework for building scalable and powerful GTs. KDLGT employs the kernel decomposition approach to rearrange the order of matrix multiplication, thereby reducing complexity to linear. Additionally, it categorizes GTs into three distinct types and provides tailored accelerating methods for each category to encompass all types of GTs. Furthermore, we provide a theoretical analysis of the performance gap between KDLGT and self-attention to ensure its effectiveness. Under this framework, we select two representative GTs to design our models. Experiments on both real-world and synthetic datasets indicate that KDLGT not only achieves state-of-the-art performance on various datasets but also reaches an acceleration ratio of approximately 10 on graphs of certain sizes.

Masked Autoencoders (MAE) have shown promising performance in self-supervised learning for both 2D and 3D computer vision. However, existing MAE-style methods can only learn from the data of a single modality, i.e., either images or point clouds, which neglect the implicit semantic and geometric correlation between 2D and 3D. In this paper, we explore how the 2D modality can benefit 3D masked autoencoding, and propose Joint-MAE, a 2D-3D joint MAE framework for self-supervised 3D point cloud pre-training. Joint-MAE randomly masks an input 3D point cloud and its projected 2D images, and then reconstructs the masked information of the two modalities. For better cross-modal interaction, we construct our JointMAE by two hierarchical 2D-3D embedding modules, a joint encoder, and a joint decoder with modal-shared and model-specific decoders. On top of this, we further introduce two cross-modal strategies to boost the 3D representation learning, which are local-aligned attention mechanisms for 2D-3D semantic cues, and a cross-reconstruction loss for 2D-3D geometric constraints. By our pre-training paradigm, Joint-MAE achieves superior performance on multiple downstream tasks, e.g., 92.4% accuracy for linear SVM on ModelNet40 and 86.07% accuracy on the hardest split of ScanObjectNN.

Melody extraction is a core task in music information retrieval, and the estimation of pitch, onset and offset are key sub-tasks in melody extraction. Existing methods have limited accuracy, and work for only one type of data, either single-pitch or multi-pitch. In this paper, we propose a highly accurate method for joint estimation of pitch, onset and offset, named JEPOO. We address the challenges of joint learning optimization and handling both single-pitch and multi-pitch data through novel model design and a new optimization technique named Pareto modulated loss with loss weight regularization. This is the first method that can accurately handle both single-pitch and multi-pitch music data, and even a mix of them. A comprehensive experimental study on a wide range of real datasets shows that JEPOO outperforms state-of-the-art methods by up to 10.6\%, 8.3\% and 10.3\% for the prediction of Pitch, Onset and Offset, respectively, and JEPOO is robust for various types of data and instruments. The ablation study validates the effectiveness of each component of JEPOO.

Point clouds obtained by LiDAR and other sensors are usually sparse and irregular. Low-quality point clouds have serious influence on the final performance of downstream tasks. Recently, a point cloud upsampling network with normalizing flows has been proposed to address this problem. However, the network heavily relies on designing specialized architectures to achieve invertibility. In this paper, we propose a novel invertible residual neural network for point cloud upsampling, called PU-INN, which allows unconstrained architectures to learn more expressive feature transformations. Then, we propose a conditional injector to improve nonlinear transformation ability of the neural network while guaranteeing invertibility. Furthermore, a lightweight interpolator is proposed based on semantic similarity distance in the latent space, which can intuitively reflect the interpolation changes in Euclidean space. Qualitative and quantitative results show that our method outperforms the state-of-the-art works in terms of distribution uniformity, proximity-to-surface accuracy, 3D reconstruction quality, and computation efficiency.

Machine learning models are incorporated in different fields and disciplines, some of which require high accountability and transparency, for example, the healthcare sector. A widely used category of explanation techniques attempts to explain models' predictions by quantifying the importance score of each input feature. However, summarizing such scores to provide human-interpretable explanations is challenging. Another category of explanation techniques focuses on learning a domain representation in terms of high-level human-understandable concepts and then utilizing them to explain predictions. These explanations are hampered by how concepts are constructed, which is not intrinsically interpretable. To this end, we propose Concept-based Local Explanations with Feedback (CLEF), a novel local model agnostic explanation framework for learning a set of high-level transparent concept definitions in high-dimensional tabular data that uses clinician-labeled concepts rather than raw features.

The last decades have witnessed significant progress in machine learning with a host of applications of algorithmic decision-making in different safety-critical domains, such as medical, law, education, and transportation. In high-stake domains, machine learning predictions have far-reaching consequences on the end-users. With the aim of applying machine learning for societal goods, there have been increasing efforts to regulate machine learning by imposing interpretability, fairness, robustness, etc. in predictions. Towards responsible and trustworthy machine learning, we propose two research themes in our dissertation research: interpretability and fairness of machine learning classifiers. In particular, we design algorithms to learn interpretable rule-based classifiers, formally verify fairness, and explain the sources of unfairness. Prior approaches to these problems are often limited by scalability, accuracy, or both. To overcome these limitations, we closely integrate automated reasoning, formal methods, and statistics with fairness and interpretability to develop scalable and accurate solutions.

We conduct a quantitative analysis of the development of the industry chain from the environmental, social, and governance (ESG) perspective, which is an overall measure of sustainability. Factors that may impact the performance of the industrial chain have been studied in the literature, such as government regulation, monetary policy, etc. Our interest lies in how the sustainability change (i.e., ESG shock) affects the performance of the industrial chain. To achieve this goal, we model the industrial chain with a graph neural network (GNN) and conduct node regression on two financial performance metrics, namely, the aggregated profitability ratios and operating margin. To quantify the effects of ESG, we propose to compute the interaction between ESG shocks and industrial chain features with a cross-attention module, and then filter the original node features in the graph regression. Experiments on two real datasets demonstrate that (i) there are significant effects of ESG shocks on the industrial chain, and (ii) model parameters including regression coefficients and the attention map can explain how ESG shocks affect the performance of the industrial chain.

Biodiversity loss is taking place at accelerated rates globally, and a business-as-usual trajectory will lead to missing internationally established conservation goals. Biosphere reserves are sites designed to be of global significance in terms of both the biodiversity within them and their potential for sustainable development, and are therefore ideal places for the development of local solutions to global challenges. While the protection of biodiversity is a primary goal of biosphere reserves, adequate information on the state and trends of biodiversity remains a critical gap for adaptive management in biosphere reserves. Passive acoustic monitoring (PAM) is an increasingly popular method for continued, reproducible, scalable, and cost-effective monitoring of animal wildlife. PAM adoption is on the rise, but its data management and analysis requirements pose a barrier for adoption for most agencies tasked with monitoring biodiversity. As an interdisciplinary team of machine learning scientists and ecologists experienced with PAM and working at biosphere reserves in marine and terrestrial ecosystems on three different continents, we report on the co-development of interactive machine learning tools for semi-automated assessment of animal wildlife.

Graph neural network (GNN) based recommender systems have become one of the mainstream trends due to the powerful learning ability from user behavior data. Understanding the user intents from behavior data is the key to recommender systems, which poses two basic requirements for GNN-based recommender systems. One is how to learn complex and diverse intents especially when the user behavior is usually inadequate in reality. The other is different behaviors have different intent distributions, so how to establish their relations for a more explainable recommender system. In this paper, we present the Intent-aware Recommendation via Disentangled Graph Contrastive Learning (IDCL), which simultaneously learns interpretable intents and behavior distributions over those intents. Specifically, we first model the user behavior data as a user-item-concept graph, and design a GNN based behavior disentangling module to learn the different intents. Then we propose the intent-wise contrastive learning to enhance the intent disentangling and meanwhile infer the behavior distributions. Finally, the coding rate reduction regularization is introduced to make the behaviors of different intents orthogonal. Extensive experiments demonstrate the effectiveness of IDCL in terms of substantial improvement and the interpretability.

Our study focuses on the United Nations Sustainable Development Goal 13: Climate Action, by identifying public attitudes on Twitter about climate change. Public consent and participation is the key factor in dealing with climate crises. However, discussions about climate change on Twitter are often influenced by the polarised beliefs that shape the discourse and divide it into communities of climate change deniers and believers. In our work, we propose a framework that helps identify different attitudes in tweets about climate change (deny, believe, ambiguous). Previous literature often lacks an efficient architecture or ignores the characteristics of climate-denier tweets. Moreover, the presence of various emotions with different levels of intensity turns out to be relevant for shaping discussions on climate change. Therefore, our paper utilizes emotion recognition and emotion intensity prediction as auxiliary tasks for our main task of stance detection. Our framework injects the words affecting the emotions embedded in the tweet to capture the overall representation of the attitude in terms of the emotions associated with it. The final task-specific and shared feature representations are fused with efficient embedding and attention techniques to detect the correct attitude of the tweet. Extensive experiments on our novel curated dataset, two publicly available climate change datasets (ClimateICWSM-2023 and ClimateStance-2022), and a benchmark dataset for stance detection (SemEval-2016) validate the effectiveness of our approach.

Due to repetitive trial-and-error style interactions between agents and a fixed traffic environment during the policy learning, existing Reinforcement Learning (RL)-based Traffic Signal Control (TSC) methods greatly suffer from long RL training time and poor adaptability of RL agents to other complex traffic environments. To address these problems, we propose a novel Adversarial Inverse Reinforcement Learning (AIRL)-based pre-training method named InitLight, which enables effective initial model generation for TSC agents. Unlike traditional RL-based TSC approaches that train a large number of agents simultaneously for a specific multi-intersection environment, InitLight pre-trains only one single initial model based on multiple single-intersection environments together with their expert trajectories. Since the reward function learned by InitLight can recover ground-truth TSC rewards for different intersections at optimality, the pre-trained agent can be deployed at intersections of any traffic environments as initial models to accelerate subsequent overall global RL training. Comprehensive experimental results show that, the initial model generated by InitLight can not only significantly accelerate the convergence with much fewer episodes, but also own superior generalization ability to accommodate various kinds of complex traffic environments.

Inferring bargainers' private valuations on items from their decisions is crucial for analyzing their strategic behaviors in bilateral sequential bargaining. Most existing approaches that infer agents' private information from observable data either rely on strong equilibrium assumptions or require a careful design of agents' behavior models. To overcome these weaknesses, we propose a Bayesian Learning-based Valuation Inference (BLUE) framework. Our key idea is to derive feasible intervals of bargainers' private valuations from their behavior data, using the fact that most bargainers do not choose strictly dominated strategies. We leverage these feasible intervals to guide our inference. Specifically, we first model each bargainer's behavior function (which maps his valuation and bargaining history to decisions) via a recurrent neural network. Second, we learn these behavior functions by utilizing a novel loss function defined based on feasible intervals. Third, we derive the posterior distributions of bargainers' valuations according to their behavior data and learned behavior functions. Moreover, we account for the heterogeneity of bargainer behaviors, and propose a clustering algorithm (K-Loss) to improve the efficiency of learning these behaviors. Experiments on both synthetic and real bargaining data show that our inference approach outperforms baselines.

In multi-agent reinforcement learning (MARL), self-interested agents attempt to establish equilibrium and achieve coordination depending on game structure. However, existing MARL approaches are mostly bound by the simultaneous actions of all agents in the Markov game (MG) framework, and few works consider the formation of equilibrium strategies via asynchronous action coordination. In view of the advantages of Stackelberg equilibrium (SE) over Nash equilibrium, we construct a spatio-temporal sequential decision-making structure derived from the MG and propose an N-level policy model based on a conditional hypernetwork shared by all agents. This approach allows for asymmetric training with symmetric execution, with each agent responding optimally conditioned on the decisions made by superior agents. Agents can learn heterogeneous SE policies while still maintaining parameter sharing, which leads to reduced cost for learning and storage and enhanced scalability as the number of agents increases. Experiments demonstrate that our method effectively converges to the SE policies in repeated matrix game scenarios, and performs admirably in immensely complex settings including cooperative tasks and mixed tasks.

Existing Unsupervised Domain Adaptation (UDA) methods typically attempt to perform knowledge transfer in a domain-invariant space explicitly or implicitly. In practice, however, the obtained features is often mixed with domain-specific information which causes performance degradation. To overcome this fundamental limitation, this article presents a novel independent feature decomposition and instance alignment method (IndUDA in short). Specifically, based on an invertible flow, we project the base features into a decomposed latent space with domain-invariant and domain-specific dimensions. To drive semantic decomposition independently, we then swap the domain-invariant part across source and target domain samples with the same category and require their inverted features are consistent in class-level with the original features. By treating domain-specific information as noise, we replace it by Gaussian noise and further regularize source model training by instance alignment, i.e., requiring the base features close to the corresponding reconstructed features, respectively. Extensive experiment results demonstrate that our method achieves state-of-the-art performance on popular UDA benchmarks. The appendix and code are available at https://github.com/ayombeach/IndUDA.

Incremental and decremental learning (IDL) deals with the tasks where new data arrives sequentially as a stream or old data turns unavailable continually due to the privacy protection. Existing IDL methods mainly focus on support vector machine and its variants with linear-type loss. There are few studies about the quadratic-type loss, whose Lagrange multipliers are unbounded and much more difficult to track. In this paper, we take the latest statistical learning framework optimal margin distribution machine (ODM) which involves a quadratic-type loss due to the optimization of margin variance, for example, and equip it with the ability to handle IDL tasks. Our proposed ID-ODM can avoid updating the Lagrange multipliers in an infinite range by determining their optimal values beforehand so as to enjoy much more efficiency. Moreover, ID-ODM is also applicable when multiple instances come and leave simultaneously. Extensive empirical studies show that ID-ODM can achieve 9.1x speedup on average with almost no generalization lost compared to retraining ODM on new data set from scratch.

We present a system for online, incremental composite event recognition. In streaming environments, the usual case is for data to arrive with a (variable) delay from, and to be revised by, the underlying sources. We propose RTEC_inc, an incremental version of RTEC, a composite event recognition engine with formal, declarative semantics, that has been shown to scale to several real-world data streams. RTEC deals with delayed arrival and revision of events by computing all queries from scratch. This is often inefficient since it results in redundant computations. Instead, RTEC_inc deals with delays and revisions in a more efficient way, by updating only the affected queries. We compare RTEC_inc and RTEC experimentally using real-world and synthetic datasets. The results are compatible with our complexity analysis and show that RTEC_inc outperforms RTEC in many practical cases.