AAAI 2025 论文列表

Achieving robust performance in vision-language tasks requires strong multimodal alignment, where textual and visual data interact seamlessly. Existing frameworks often combine contrastive learning with image captioning to unify visual and textual representations. However, reliance on global representations and unidirectional information flow from images to text limits their ability to reconstruct visual content accurately from textual descriptions. To address this limitation, we propose BiMAC, a novel framework that enables bidirectional interactions between images and text at both global and local levels. BiMAC employs advanced components to simultaneously reconstruct visual content from textual cues and generate textual descriptions guided by visual features. By integrating a text-region alignment mechanism, BiMAC identifies and selects relevant image patches for precise cross-modal interaction, reducing information noise and enhancing mapping accuracy. BiMAC achieves state-of-the-art performance across diverse vision-language tasks, including image-text retrieval, captioning, and classification.

Complex claim fact-checking performs a crucial role in disinformation detection. However, existing fact-checking methods struggle with claim vagueness, specifically in effectively handling latent information and complex relations within claims. Moreover, evidence redundancy, where non-essential information complicates the verification process, remains a significant issue. To tackle these limitations, we propose Bilateral Defusing Verification (BiDeV), a novel fact-checking working-flow framework integrating multiple role-played LLMs to mimic the human-expert fact-checking process. BiDeV consists of two main modules: Vagueness Defusing identifies latent information and resolves complex relations to simplify the claim, and Redundancy Defusing eliminates redundant content to enhance the evidence quality. Extensive experimental results on two widely used challenging fact-checking benchmarks (Hover and Feverous-s) demonstrate that our BiDeV can achieve the best performance under both gold and open settings. This highlights the effectiveness of BiDeV in handling complex claims and ensuring precise fact-checking.

Molecular representation learning is vital for various downstream applications, including the analysis and prediction of molecular properties and side effects. While Graph Neural Networks (GNNs) have been a popular framework for modeling molecular data, they often struggle to capture the full complexity of molecular representations. In this paper, we introduce a novel method called Gode, which accounts for the dual-level structure inherent in molecules. Molecules possess an intrinsic graph structure and simultaneously function as nodes within a broader molecular knowledge graph. Gode integrates individual molecular graph representations with multi-domain biochemical data from knowledge graphs. By pre-training two GNNs on different graph structures and employing contrastive learning, Gode effectively fuses molecular structures with their corresponding knowledge graph substructures. This fusion yields a more robust and informative representation, enhancing molecular property predictions by leveraging both chemical and biological information. When fine-tuned across 11 chemical property tasks, our model significantly outperforms existing benchmarks, achieving an average ROC-AUC improvement of 12.7% for classification tasks and an average RMSE/MAE improvement of 34.4% for regression tasks. Notably, Gode surpasses the current leading model in property prediction, with advancements of 2.2% in classification and 7.2% in regression tasks.

Semi-supervised learning (SSL) is a fundamental task in machine learning, empowering models to extract valuable insights from datasets with limited labeled samples and a large amount of unlabeled data. Although pseudo-labeling is a widely used approach for SSL that generates pseudo-labels for unlabeled data and leverages them as ground truth labels for training, traditional pseudo-labeling techniques often face challenges that significantly decrease the quality of pseudo-labels and hence the overall model performance. In this paper, we propose a novel Bi-level Optimization method for Pseudo-label Learning (BOPL) to boost semi-supervised training. It treats pseudo-labels as latent variables, and optimizes the model parameters and pseudo-labels jointly within a bi-level optimization framework. By enabling direct optimization over the pseudo-labels towards maximizing the prediction model performance, the method is expected to produce high-quality pseudo-labels. To evaluate the effectiveness of the proposed approach, we conduct extensive experiments on multiple SSL benchmarks. The experimental results show the proposed BOPL outperforms the state-of-the-art SSL techniques.

Dataset condensation has significantly improved model training efficiency, but its application on devices with different computing power brings new requirements for different data sizes. For sparse graph data with non-Euclidean structures, repeated condensation of each scale may lead to significant computational costs. Thus, condensing multiple scale graphs simultaneously is the core of achieving efficient training in different on-device scenarios. Existing efficient works for multi-scale graph dataset condensation mainly perform efficient approximate computation in scale order (large-to-small or small-to-large scales). However, these two commonly used paradigms for multi-scale graph dataset condensation have serious ''scaling down degradation'' and ''scaling up collapse" problems of a graph. The main bottleneck of the above paradigms is whether the effective information of the original graph is fully preserved when consenting to the primary sub-scale (the first of multiple scales), which determines the condensation effect and consistency of all scales. In this paper, we proposed a novel GNN-centric Bi-directional Multi-Scale Graph Dataset Condensation (BiMSGC) framework, to explore unifying paradigms by operating on both large-to-small and small-to-large for multi-scale graph condensation. Based on the mutual information theory, we estimate an optimal ''meso-scale'' to obtain the minimum necessary dense graph preserving the maximum utility information of the original graph, and then we achieve stable and consistent ''bi-directional'' condensation learning by optimizing graph eigenbasis matching with information bottleneck on other scales. Encouraging empirical results on several datasets demonstrates the significant superiority of the proposed framework in graph condensation at different scales.

Sexism affects both women and men, yet research often overlooks misandry and suffers from overly broad annotations that limit AI applications. To address this, we introduce BeyondGender, a dataset meticulously annotated according to the latest definitions of misogyny and misandry. It features innovative multifaceted labels encompassing aspects of sexism, gender, phrasing, misogyny, and misandry. The dataset includes 6K English and 1.7K Chinese sexism instances, alongside 13K non-sexism examples. Our evaluations of masked language models and large language models reveal that they detect misogyny in English and misandry in Chinese more effectively, with F1-scores of 0.87 and 0.62, respectively. However, they frequently misclassify hostile and mild comments, underscoring the complexity of sexism detection. Parallel corpus experiments suggest promising data augmentation strategies to enhance AI systems for nuanced sexism detection, and our dataset can be leveraged to improve value alignment in large language models.

The task of 3D object detection is crucial for various applications that rely on identifying objects in three-dimensional space using inputs like LiDAR point clouds and images. However, LiDAR-based detection faces challenges due to the sparsity of point clouds, especially at greater distances. To address this, depth completion models have been used to generate virtual points from RGB images, but they struggle with real-time applications due to high computational costs. Our work eliminates the depth completion process, significantly improving processing speed while minimizing performance degradation. Consequently, our method has achieved an optimal balance between speed and accuracy on the KITTI leaderboard.

Fact verification has become increasingly vital in the internet age, driven by the proliferation of false claims and political misinformation. While traditional methods rely predominantly on text-based evidence, multi-modal evidence introduces richer sources of information, offering valuable insights for claim verification. Existing multi-modal verification models often focus on superficial correlations between claims and evidence, neglecting the complex semantic interactions present in fine-grained multi-modal signals. In this paper, we propose a novel framework for multi-modal fact-checking, named Hypergraph Transformer-based Multi-modal Fact-Checking (HGTMFC). Our approach captures high-order relationships between different modalities of evidence and claims by leveraging hypergraphs. HGTMFC models the intricate relationships among evidence across various modalities and enhances information propagation through a transformer-based mechanism embedded within the hypergraph. Moreover, we utilize linegraphs to refine this propagation process, further strengthening the model's reasoning capabilities. Experiments on benchmark datasets demonstrate that our model significantly outperforms existing approaches in multi-modal fact verification.

Vanilla convolution and window-based self-attention have shown significant success in image dehazing. However, they are constrained by limited receptive fields and ignore frequency gaps between dehazed and clear images. The former hampers the modeling of global dependencies, while the latter impedes the learning of high-frequency features, leading to suboptimal performance. In this paper, we propose the Joint Spatial and Fourier Convolutional Network (JSFC-Net), which leverages Fourier transformation to simultaneously address the two aforementioned problems with low computational overhead. We introduce the Frequency-Spatial Promoted and Physical Learning Block, which extracts high-level features from the spatial domain and frequency domain in parallel. We design a simple yet effective solution that uses spatial features to promote and modulate frequency features in a multi-scale manner, achieving refinement of frequency features and addressing robustness issue caused by global sensitivity. Additionally, we present the Receptive Field Selection Module to facilitate improved fusion of spatial and frequency domain features. Finally, we introduce frequency loss to further narrow frequency gaps. Comprehensive experiments on multiple datasets demonstrate that JSFC-Net is significantly superior to SOTA dehazing methods.

Training deep Convolutional Neural Networks (CNNs) presents unique challenges, including the pervasive issue of elimination singularities—consistent deactivation of nodes leading to degenerate manifolds within the loss landscape. These singularities impede efficient learning by disrupting feature propagation. To mitigate this, we introduce Pool Skip, an architectural enhancement that strategically combines a Max Pooling, a Max Unpooling, a 3 × 3 convolution, and a skip connection. This configuration helps stabilize the training process and maintain feature integrity across layers. We also propose the Weight Inertia hypothesis, which underpins the development of Pool Skip, providing theoretical insights into mitigating degradation caused by elimination singularities through dimensional and affine compensation. We evaluate our method on a variety of benchmarks, focusing on both 2D natural and 3D medical imaging applications, including tasks such as classification and segmentation. Our findings highlight Pool Skip's effectiveness in facilitating more robust CNN training and improving model performance.

Emotion recognition in conversation (ERC) has been promoted with diverse approaches in the recent years. However, many studies have pointed out that emotion shift and confusing labels make it difficult for models to distinguish between different emotions. Existing ERC models suffer from these problems when the emotions are forced to be mapped into single label. In this paper, we utilize our strategies for extending single label to multi-labels. We then propose a multi-label classification framework for emotion recognition in conversation (ML-ERC). Specifically, we introduce weighted supervised contrastive learning tailored for multi-label, which can easily applied to previous ERC models. The empirical results on existing task with single label support the efficacy of our approach, which is more effective in the most challenging settings: emotion shift or confusing labels. We also evaluate ML-ERC with the multi-labels we produced to support our contrastive learning scheme.

In the rapidly developing field of automatic text generation and understanding, the quality of input data has been shown to be a key factor affecting the efficiency and accuracy of large language model (LLM) output. With the advent of advanced tools such as ChatGPT, input refinement work has mainly focused on prompt engineering. However, existing methods are often too dependent on specific contexts and are easily affected by individual expert experience and potential biases, limiting their wide applicability in diverse real-world applications. To address this problem, this study develops an Reinforced Token-Level Input Refinement, called RTLIR. We choose to optimize the input data at the fine-grained level of tokens, cleverly preserving the original text structure. Operationally, each state is defined by the token set of the current text, and each action is a binary decision process to decide whether to retain a specific token information. The agent automatically calculates and determines the selection probability of each token based on the current state, thereby optimizing the entire decision process. Through continuous exploration and learning, the agent can autonomously learn to identify the key inputs that have the greatest impact on the generation results and achieve refinement of the input data. In addition, RTLIR is a plug-and-play, LLM-agnostic module that can be used for a wide range of tasks and models. Experimental results show that RTLIR improves the performance of LLM in various input scenarios and tasks, with an average accuracy increase of 6%.

Few-Shot Video Object Segmentation (FSVOS) aims to achieve accurate segmentation of video sequences supported by limited annotated images. In this work, we analyze the deficiencies inherent in the use of object prototypes and pixel features as references in previous methods. Then we shed light on that part features, with the ability to adapt to appearance variations and resist noise, are advantageous as representative reference features for aligning support images and query videos. Therefore, we propose a Part Agent Learning Network (PALN) to leverage part features from two aspects. First, we elaborately employ Optimal Transport algorithm with equal partition constraint to make part agents capable of dividing support objects into diverse parts in an adaptive manner. Second, we design a dedicated cache mechanism to learn temporal part agents as lightweight historic target representation to exploit temporal consistency. With the aid of these learned part agents, our PALN can effectively achieve support-query alignment and temporal alignment for accurate segmentation of query videos. Extensive experimental results on two challenging benchmarks demonstrate that our method performs favorably against state-of-the-art FSVOS methods.

Equilibrium problems in Bayesian auction games can be described as systems of differential equations. Depending on the model assumptions, these equations might be such that we do not have a rigorous mathematical solution theory. The lack of analytical or numerical techniques with guaranteed convergence for the equilibrium problem has plagued the field and limited equilibrium analysis to rather simple auction models such as single-object auctions. Recent advances in equilibrium learning led to algorithms that find equilibrium under a wide variety of model assumptions. Monotonicity and the Minty condition are the known sufficient conditions for learning algorithms to converge to an equilibrium in games. Not much is known about convergence of learning algorithms beyond these conditions. We analyze first- and second-price auctions where simple learning algorithms consistently converge to an equilibrium. The analysis is challenging, because these properties need to be shown in infinite dimensions. Interestingly, we show that neither monotonicity nor pseudo- or quasi-monotonicity holds for the respective variational inequalities (VIs). The second-price auction's equilibrium is a Minty-type solution, but the first-price auction is not. However, the analysis via infinite-dimensional VIs allows us to get ex-post guarantees for gradient-based algorithms. We show that the Bayes--Nash equilibrium is the unique solution to the VI within the class of uniformly increasing bid functions, which ensures that gradient-based algorithms attain the equilibrium in case of convergence, as also observed in numerical experiments.

In the field of mixed-motive games, extensive multi-agent learning studies have explored the balance between egoism (individual interest), utilitarianism (collective interest), and egalitarianism (fairness). Traditional approaches often rely on manually designed reward functions, social norms, and alliance/federation mechanisms to transition agents from individualistic behaviors toward cooperative strategies. However, these methods typically require all agents to share private local information or to mandatorily participate in federations, which is impractical in real-world applications. To address these issues, this paper proposes a Flexible-Participation Federation (FPF) framework that allows agents to participate in the federation voluntarily. Furthermore, we extend the federation from a global to a Local Multi-Federation (LMF) framework, enabling agents to form multiple localized federations, thereby promoting more efficient and adaptive cooperation. Theoretical evidence demonstrates that the global FPF model, along with the discrepancy between decentralized egoistic policies and federated utilitarian policies, achieves an O(1/T) convergence rate. Agents in the LMF framework also reach consensus within a sublinear gap. Extensive experiments show that agents opting out of federation participation experience a reduction in egoism, and our approach outperforms multiple baselines in terms of both utilitarianism and egalitarianism.

With the availability of various instruction datasets, a pivotal challenge is how to effectively select and integrate these instructions to fine-tune large language models (LLMs). Previous research mainly focuses on selecting individual high-quality instructions. However, these works overlooked the joint interactions and dependencies between different categories of instructions, leading to suboptimal selection strategies. Moreover, the nature of these interaction patterns remains largely unexplored, let alone optimize the instruction set with regard to them. To fill these gaps, in this paper, we: (1) systemically investigate interaction and dependency patterns between different categories of instructions, (2) manage to optimize the instruction set concerning the interaction patterns using a linear programming-based method, and optimize the learning schema of SFT using an instruction dependency taxonomy guided curriculum learning. Experimental results across different LLMs demonstrate improved performance over strong baselines on widely adopted benchmarks.

Human preference alignment can significantly enhance the capabilities of Multimodal Large Language Models (MLLMs). However, collecting high-quality preference data remains costly. One promising solution is the self-evolution strategy, where models are iteratively trained on data they generate. Current multimodal self-evolution techniques, nevertheless, still need human- or GPT-annotated data. Some methods even require extra models or ground truth answers to construct preference data. To overcome these limitations, we propose a novel multimodal self-evolution framework that empowers the model to autonomously generate high-quality questions and answers using only unannotated images. First, in the question generation phase, we implement an image-driven self-questioning mechanism. This approach allows the model to create questions and evaluate their relevance and answerability based on the image content. If a question is deemed irrelevant or unanswerable, the model regenerates it to ensure alignment with the image. This process establishes a solid foundation for subsequent answer generation and optimization. Second, while generating answers, we design an answer self-enhancement technique to boost the discriminative power of answers. We begin by captioning the images and then use the descriptions to enhance the generated answers. Additionally, we utilize corrupted images to generate rejected answers, thereby forming distinct preference pairs for effective optimization. Finally, in the optimization step, we incorporate an image content alignment loss function alongside the Direct Preference Optimization (DPO) loss to mitigate hallucinations. This function maximizes the likelihood of the above generated descriptions in order to constrain the model's attention to the image content. As a result, model can generate more accurate and reliable outputs. Experiments demonstrate that our framework is competitively compared with previous methods that utilize external information, paving the way for more efficient and scalable MLLMs.

Graph contrastive learning (GCL) has drawn much research attention for its ability to learn node representations in a self-supervised manner. However, the homophily assumption inherent in GNN encoders limits the direction (macro-level) and the process (micro-level) of message passing in current GCL frameworks, impairing the expressive power of GCL in non-homophilous graphs. This paper presents a novel framework that employs Macro and Micro Message Passing in GCL (M3P-GCL) to overcome these limitations and advance performance in both homophilous and non-homophilous graphs. Specifically, at the macro-level, we integrate structural and attribute views to enhance the direction of message passing, and employ an Aligned Priority-Supporting View Encoding (APS-VE) strategy to facilitate contrastive training; at the micro-level, we propose an Adaptive Self-Propagation (ASP) strategy based on role segmentation of self-loops to diversify the process of message passing in the encoder. These enhancements effectively address the limitations imposed by the homophily assumption. Experiments demonstrate that M3P-GCL outperforms both supervised and unsupervised baselines in the node classification task on various datasets with different levels of homophily.

Given the large volume of side information from different modalities, multimodal recommender systems have become increasingly vital, as they exploit richer semantic information beyond user-item interactions. Recent works highlight that leveraging Graph Convolutional Networks (GCNs) to explicitly model multimodal item-item relations can significantly enhance recommendation performance. However, due to the inherent over-smoothing issue of GCNs, existing models benefit only from shallow GCNs with limited representation power. This drawback is especially pronounced when facing complex and high-dimensional patterns such as multimodal data, as it requires large-capacity models to accommodate complicated correlations. To this end, in this paper, we investigate bypassing GCNs when modeling multimodal item-item relationship. More specifically, we propose a Topology-aware Multi-Layer Perceptron (TMLP), which uses MLPs instead of GCNs to model the relationships between items. TMLP enhances MLPs with topological pruning to denoise item-item relations and intra (inter)-modality learning to integrate higher-order modality correlations. Extensive experiments on three real-world datasets verify TMLP's superiority over nine baselines. We also find that by discarding the internal message passing in GCNs, which is sensitive to node connections, TMLP achieves significant improvements in both training efficiency and robustness against existing models.