IJCAI 2026 论文列表

Image clustering aims to partition unlabeled image datasets into distinct groups. A core aspect of this task is constructing and leveraging prior knowledge to guide the clustering process. Recent approaches introduce semantic descriptions as prior information, most of which typically relying on matching-based techniques with predefined vocabularies. However, the limited matching space restricts their adaptability to downstream clustering tasks. Moreover, these methods primarily focus on reducing bias to improve performance, frequently overlooking the importance of variance reduction. To address these limitations, we propose GSEC (Image Clustering based on Generative Semantic Guidance and Bi-Layer Ensemble), a framework designed to reduce bias through generative semantic guidance and mitigate variance via ensemble learning. Our method employs Multimodal Large Language Models to generate semantic descriptions and derive image embeddings via weighted averaging. Additionally, a bi-layer ensemble strategy integrates cross-modal information through BatchEnsemble in the inner layer and aligns outputs via an alignment mechanism in the outer layer. Comparative experiments demonstrate that GSEC outperforms 20 state-of-the-art methods across six benchmark datasets, while further analysis confirms its effectiveness in simultaneously reducing both bias and variance.

Split Federated Learning (SFL) has emerged as a pivotal paradigm for privacy-preserving distributed training on resource-constrained edge devices by partitioning neural networks between clients and a server. A critical design choice in SFL is the split layer, which determines the computation distribution and the semantic level of smashed data, directly impacting communication overhead, training latency, and model accuracy. Existing SFL methods largely ignore the time-varying nature of edge resources, relying on static resource profiles that lead to suboptimal efficiency and compromise model convergence. In this paper, we propose Dynamic Adaptive Split Federated Learning (DASFL), a unified framework that jointly addresses dynamic resource heterogeneity and non-IID data distributions. We employ a resource-aware dynamic split layer selection strategy that enables each client to minimize per-round latency by adapting to instantaneous local conditions. We also design an adaptive masked aggregation that robustly synchronizes client updates under split-induced structural heterogeneity. Extensive experiments on multiple benchmark models and datasets, conducted under heterogeneous, time-varying resource conditions and non-IID data distributions, demonstrate that DASFL achieves a superior accuracy-efficiency trade-off and faster convergence compared to state-of-the-art SFL baselines.

Online planning in continuous state, action, and observation spaces remains challenging for autonomous systems. While Monte Carlo Tree Search (MCTS) scales effectively via sampling, most continuous (PO)MDP solvers do not exploit gradient-based action optimization. We propose Action-Gradient MCTS (AGMCTS), a framework that combines global tree search with local gradient-based action refinement, while maintaining consistent value estimates. We provide three key theoretical contributions: (1) an action score gradient theorem for particle belief states; (2) the Multiple Importance Sampling (MIS) Tree that supports frequent action-branch updates by reusing prior samples without introducing estimator drift; and (3) tractable action score gradients for smooth generative models using the Area Formula. Empirical results demonstrate that AGMCTS outperforms state-of-the-art sample-based solvers in multiple challenging continuous MDP and POMDP benchmarks.

The learnware paradigm supports model reuse by pairing each submitted model with a specification, a lightweight representation used by the learnware dock system to identify, match, and reuse models without accessing raw data. While specifications are essential for learnware identification, they are also data-dependent public artifacts and it is not clear whether they reveal private information. Recently, the Parameter Vector (PAVE) specification has been proposed and shown to be effective for learnwares, yet its privacy properties remain largely unexplored. In this paper, we provide the first theoretical privacy analysis for PAVE. Specifically, we first formalize two specification-induced risks in the learnware paradigm: the disclosure risk of the released specification and the amplification risk that the specification may strengthen attacks against the released model. Second, we characterize when compact PAVE releases admit intrinsic differential privacy (DP): under natural structural conditions of learnware docks, the compact PAVE specification satisfies an (ε, δ)-DP guarantee without explicit additive noise through a Gaussian-sketch view of stable parameter variations, and for regimes outside these conditions, we further provide DP-Stabilized-PAVE as a certified differentially private variant. Third, we show that the resulting DP guarantees control both disclosure risk and amplification risk, and we analyze the induced privacy--utility trade-off to guide effective learnware identification while preserving privacy.

Discovering shapelets -- i.e., discriminative temporal patterns within time series -- has been widely studied to address the inherent complexity of time-series classification (TSC) and to make model decision-making processes more transparent. However, existing methods primarily focus on population-level shapelets optimized across the entire dataset, which leads to two fundamental limitations: (i) population-level patterns often misalign with instance-specific features, resulting in suboptimal performance and potentially misleading interpretations, and (ii) most methods treat shapelets as independent entities, overlooking important temporal dependencies and interactions among multiple patterns. To address these limitations, we propose INSHAPE, an interpretable TSC framework that discovers variable-length, discriminative temporal patterns specific to each time series. INSHAPE identifies these patterns as non-overlapping segments and models their temporal dependencies, thereby providing clear instance-level interpretations while achieving strong predictive performance. Furthermore, INSHAPE bridges local and global interpretability through a bottom-up approach, aggregating instance-level shapelets into prototypical (population-level) shapelets. Extensive experiments on 128 UCR and 30 UEA benchmark datasets show that INSHAPE consistently outperforms state-of-the-art shapelet-based methods while providing more intuitive and interpretable insights.

In many multi-agent applications, tasks yield rewards only when executed by a coalition meeting an unknown size threshold; otherwise, feedback is fully censored. This censorship creates an identifiability problem: agents cannot distinguish stochastic failure from insufficient coordination. We formalize this setting as the Threshold-Activated Cooperative Multi-Armed Bandit (TAC-MAB) and analyze it under both centralized and decentralized coordination. We show that a centralized algorithm (C-TAC) achieves cumulative regret O(log T), decomposed into a structural-search term that captures the cost of resolving feasibility under censored feedback and a statistical-monitoring term for value estimation. We then introduce D-TAC, a decentralized event-triggered protocol in which agents synchronize only when their structural beliefs change. Empirically, D-TAC achieves a 23x reduction in communication relative to the centralized baseline while preserving feasibility alignment under conservative belief fusion. These results characterize the coordination cost of learning under censored feedback and show that near-centralized communication efficiency is achievable without continuous synchronization.

Driven by privacy regulations, federated unlearning (FU) aims to remove the influence of specific clients or samples from a trained federated model, approximating the behavior of retraining from scratch without the target data. However, existing FU methods are largely reactive: retraining-based solutions are accurate but prohibitively expensive, while parameter-based approaches are more efficient yet may cause irreversible knowledge damage and catastrophic forgetting. We introduce PFU-SKLA, a proactive FU framework that endows models with built-in forgettability. During pretraining, we perform Orthogonality-guided Representation Disentanglement (ORD) to learn a robust, disentangled feature space that reduces cross-client/class interference. Additionally, we use Dynamic Sensitivity-based Key Layer Identification (DS-KLI) to identify sensitive layers that hold target knowledge. During unlearning, we propose a sparse, lightweight adaptation strategy that precisely erases target knowledge by inserting sparse adapters into the identified critical layers while preserving non-target knowledge by freezing the backbone. Extensive experiments across multiple datasets and three standard unlearning settings show that PFU-SKLA consistently approaches retraining-from-scratch performance, while substantially reducing communication and computation costs compared to state-of-the-art (SOTA) FU methods.

End-to-end autonomous driving systems have demonstrated advantages over traditional modular systems. Despite this progress, these end-to-end systems still struggle to be deployed in real-world driving environments, as they inevitably encounter undertrained scenarios in which autonomous vehicles may take unsafe actions. Reinforcement Learning (RL) provides a theoretical framework for addressing this challenge by enabling autonomous vehicles to self-improve: continuously collecting additional scenarios and learning from them. However, training autonomous vehicles with RL is not straightforward in the real world. Collecting real-world driving data involves costly interactions with the environment, and significant human intervention is required both to prevent autonomous vehicles from entering unsafe states and to reset them for subsequent episodes. In this paper, we introduce a novel real-world RL algorithm that allows autonomous vehicles to collect informative scenarios and learn from them with minimal human intervention. Our algorithm considers the learning progress of autonomous vehicles to identify informative scenarios and abort episodes before they enter unsafe states. To evaluate our algorithm, we introduce challenging urban driving tasks that require autonomous vehicles to reset themselves to initial states. The experimental results show that our real-world RL algorithm outperforms baselines with much less human intervention.

Predicting cross-sectional stock returns is challenging due to low signal-to-noise ratios and evolving market regimes. Classical factor models offer interpretability but limited flexibility, while deep learning models achieve strong performance yet often underutilize financial priors. We address this gap with PRISM-VQ (PRior-Informed Stock Model with Vector Quantization), a dynamic factor framework that integrates expert prior factors, vector-quantized discrete latent factors learned from cross-sectional structure, and a structure-conditioned Mixture-of-Experts to generate time-varying factor loadings. Vector quantization acts as an information bottleneck that suppresses noise while capturing robust market structure, with discrete codes serving both as latent factors and as routing signals for temporal expert specialization. Experiments on CSI 300 and S&P 500 show consistent improvements in cross-sectional return prediction and portfolio performance over strong baselines while preserving interpretability. Our code is available at https://github.com/finxlab/PRISM-VQ.

Label Distribution Learning (LDL) effectively addresses label ambiguity by modeling the degree to which each label describes an instance. A key challenge in LDL is Label Enhancement (LE): recovering label distributions from logical labels. Existing LE methods typically treat logical labels as supervisory signals and learn a direct mapping from features to label distributions. However, they fail to fully exploit the rich information encoded in logical labels, limiting their performance. We propose CVMG (Cross-View Fusion and Mixed Graph-based Label Enhancement), a novel approach that addresses this limitation through two key innovations. First, we employ a cross-attention mechanism to integrate logical labels and features, leveraging their complementary information to generate enriched feature representations. Second, we construct a mixed dependency graph that captures both instance-level relationships from enhanced features and category-level dependencies from logical labels. Label distributions are then recovered through propagation over this graph. Extensive experiments on 13 real-world datasets demonstrate that CVMG significantly outperforms state-of-the-art methods, validating the effectiveness of our approach.

Training with instance-dependent label noise poses a critical challenge in deep learning, severely impeding model generalization. Existing methods often struggle to break the vicious cycle where corrupted labels degrade feature representations, which in turn impairs noise identification. To overcome this, we propose the Meta-Cognitive Resonance Label Correction (MecReC) framework. For feature representation, MecReC structurally decouples representation learning from label correction, leveraging a pre-trained model as a frozen anchor to ensure feature robustness. For label correction, we introduce a novel meta-cognitive resonance mechanism that aggregates multiple complementary signals to comprehensively assess label reliability. These collaborative signals are fed to a lightweight meta-weight network, which automatically learns whether and how to correct each label. Furthermore, our personalized confidence thresholding strategy guarantees a tighter error bound for label purification. Extensive experiments demonstrate that MecReC consistently outperforms SOTA methods, exhibiting superior robustness particularly under extreme noise conditions. Code and additional details are available at https://github.com/JieSu272/MecReC.

Deep incomplete multi-view clustering methods can mine patterns of incomplete multi-view data without labels, gaining great attention in various domains. However, current methods are obsessed with aligning view-specific representations from available samples with complete views to learn view-invariant information for missing data recovery, ignoring fruitful view-complementary information beneficial to data imputations. Meanwhile, such the view-invariant learning often leads to the problem of dominant view dependency that models tend to over-rely on views with clear clustering structures and neglect weaker ones, causing a performance bottleneck. Therefore, a content-style disentanglement guided representation learning is proposed for the incomplete multi-view clustering (CSMVC). Specifically, CSMVC designs a view-specific dual-representation learning architecture that extracts view-invariant content representations and view-specific style representations from self-supervised reconstructions of each view with the guidance of the Hilbert schmidt independence criterion. Then, it performs a content-centered style-modulated representation imputation mechanism to infer missing data via fully utilizing inter-view complementary and consistent information. Meanwhile, it devises a cross-view consensus structure mining strategy to capture the clustering structure with intra-cluster compactness and inter-cluster separability from attention-based fusion representations that adaptively aggregate content representations and style representations of each view. Finally, comprehensive experiments show that CSMVC outperforms state-of-the-art methods.

Spatial transcriptomics has significantly advanced tissue biology and makes it possible to study the spatial interactions of cells in the microenvironment of complex tissues. However, accurately inferring intercellular communication from these data remains challenging due to the need to effectively integrate spatial topology with high-dimensional gene expression information. To address this challenge, we propose DAHGT-CCI, a novel method for inferring intercellular communication from spatial transcriptomics data. By fusing multimodal heterogeneous features to construct a unified node representation and introducing a meta-relation-aware dynamic attention mechanism, DAHGT-CCI can adaptively learn the weights of edges in the graph, thereby achieving superior precision and fine-grained inference of intercellular interactions. We evaluated the performance of DAHGT-CCI on six different spatial transcriptomics datasets. In these benchmarks, DAHGT-CCI outperformed existing graph-based learning methods and cell-cell communication inference models in terms of accuracy, AUROC, recall, and F1 score. These results demonstrate that DAHGT-CCI can more accurately reconstruct cell communication networks in complex tissue microenvironments, offering an indispensable computational tool for studying developmental processes, disease mechanisms, and potential therapeutic targets from a spatially resolved perspective.

The scarcity of paired gene expression and pathology images datasets poses a major bottleneck for training large-scale pathology foundation models. Although gene-to-image generative models offer a promising solution, existing methods typically employ coarse-grained conditional control strategies, resulting in entanglement between background textures and cell layouts. To address this challenge, we propose GeneCaDiff, a three-stage cascaded diffusion model that explicitly aligns the generative process with the hierarchical organization of biological tissues. Two conditional DDPMs independently synthesize tissue backgrounds and cellular foreground components, and a subsequent ControlNet-based fusion generator utilizes niche and cellular community maps as dual conditions to synthesize realistic tissue images. Quantitative and qualitative evaluations demonstrate strong realism and diversity, while controllability experiments validate hierarchical, decoupled control over niche textures and community layouts.

Existing multi-modal automatic modulation recognition (AMR) methods primarily focus on exploiting multi-view representations of raw signal data to improve performance, but still struggle to effectively model and exploit high-level human prior knowledge. Although recent studies attempt to introduce large language models (LLMs) to integrate the textual modality, most of them merely treat LLMs as feature extractors, without further exploiting the LLM's potential. To this end, we propose a knowledge-enhanced multi-modal automatic modulation recognition framework based on large language model (AMR-LLM). Specifically, we first specially design an AMR instruction construction mechanism based on knowledge to activate the LLM's potential for signal perception, which designates the LLM as a signal domain expert, introduces human prior knowledge as a supplement, and exploits the unique physical information of each data sample as connection guidance. Then, to enable the LLM to directly perceive digital signals, we introduce a progressive multi-modal fusion strategy mapping signal features into the space of LLMs. Experimental results on multiple benchmark datasets demonstrate that AMR-LLM achieves approximately 5% higher accuracy than current state-of-the-art methods. Moreover, it achieves the current domain performance with only 30% of the data.

As a fundamental component of the physical layer, convolutional codes are instrumental in ensuring dependable transmission of information over inherently noisy communication channels. Recently, neural network-based decoding algorithms have shown remarkable progress in enhancing the performance of block error-correcting codes. However, current neural network decoders for convolutional codes often exhibit strong performance primarily on low-rate codewords, whereas maximum likelihood decoding algorithm incur substantial time overhead when decoding high-rate codewords, which impacts communication efficiency in practical communication systems. In this work, we propose a Transformer-based decoder for convolutional codes designed for soft-decision decoding over AWGN and fading channels, which serves as an efficient decoding scheme for convolutional codes with various code rates. For the unique structure of convolutional codes, we propose a Bidirectional Context Aggregation mechanism to enable the model to fully capture the long-range dependencies and the memory effect of convolutional structures. Additionally, we extend the Transformer attention matrix with a codeword-specific feature integration mechanism to allow the model to capture implicit relationships within codewords. Extensive experiments demonstrate that the accuracy performance of the proposed model outperforms that of neural network decoders. And for maximum likelihood decoding algorithm, our model achieves comparable error-correction capability and a significantly faster decoding speed.

Decision-making pipelines increasingly rely on prediction models whose outputs serve as inputs to downstream optimization problems. Decision-Focused Learning (DFL) has emerged as a promising approach to training such models by directly optimizing decision quality rather than predictive accuracy alone. While most existing DFL methods assume a complete-information setting in which the ground-truth optimization parameters are observed, this paper studies Contextual Inverse Optimization (CIO), an incomplete-information setting in which only the resulting solutions are observed. Prior work on CIO has proposed learning algorithms based on optimality conditions for linear programs, as well as methods that repeatedly solve inverse optimization problems to handle integer programs, often incurring a substantial computational burden. In this paper, we propose a learning algorithm for general optimization problems with linear objective functions that eliminates the need to solve inverse optimization problems. The proposed method learns prediction models by solving a Relaxed Inverse Optimization Problem (RIOP), constructed based on feasible solutions randomly sampled from the feasible region, thereby reducing the computational overhead associated with existing CIO methods. Numerical experiments demonstrate that our method achieves competitive performance in terms of regret compared with existing methods, while offering improved computational efficiency for certain classes of downstream optimization problems.

To make optimal joint pricing and inventory control decisions is a critical challenge for modern retailers. In practice, retailers face changing market conditions where demands are influenced by various contextual factors, while simultaneously dealing with the difficulty of lost sales that obscure true demand information. However, existing approaches often fail to account for both contextual information and censored demand observations. We address this gap by presenting a framework where we model demand as a linear combination of basis functions with unknown coefficients, allowing for adaptive pricing and inventory decisions that respond to changing contexts. We propose an efficient algorithm to achieve regret bound O(K sqrt(T) log T) under concave revenue conditions and O(K^(2/3) T^(2/3) (log T)^(1/2)) for the general case, with matching lower bounds confirming optimality. Extensive numerical experiments across diverse scenarios demonstrate our algorithm’s effectiveness.

Federated knowledge distillation (Fed-KD) exchanges distilled predictions on a shared proxy dataset, often reducing communication and accommodating heterogeneous client architectures. However, in multimodal federated learning (MFL), modality dominance can bias local optimization and contaminate the aggregated global teacher, degrading both multimodal accuracy and robustness to missing modalities, especially under non-IID heterogeneity. We propose Federated Adaptive-Modality Balancing (FedAMB). At the client side, Selective-Modality Regulation (SMR) models dominance as a state-dependent phenomenon and intervenes only when it destabilizes training, strengthening weak modalities without over-regularization. At the server side, Component-wise Modality Distillation (CMD) regulates how aggregated knowledge is transferred to each modality branch, preventing the propagation of fusion-biased teachers while directly improving unimodal representations. Experiments on CREMA-D, AVE, and UR-FUNNY show that FedAMB consistently improves multimodal accuracy and missing-modality robustness.