IJCAI 2026 Papers

Dynamic link prediction aims to predict whether two nodes will interact at a future time point in a dynamic graph based on their historical interactions. Existing sampling based methods, which can be considered as a compressor, generally select a subset of one-hop neighbors from the entire interaction sequence for efficiency. It inevitably makes them lose rich context such as high order structural information and temporal patterns, and often underperform on nodes with sparse historical interaction with other nodes. To this aim, we propose a novel context construction based approach named PeNS that adopts Pseudo Neighbor Augmentation Sampling Strategy for more accurate dynamic link prediction. Specifically, PeNS tries to augment neighbor sequences with three complementary types of pseudo neighbors, injecting structural and temporal cues to provide rich context information especially in sparse neighbor scenarios. We further introduce a Streaming Neighbor Memory Module for efficient pseudo neighbor construction and a dual-stream fusion mechanism for robust representation learning. Extensive experiments demonstrate the effectiveness and generalizability of PeNS.

Backdoor attacks pose a serious threat to deep reinforcement learning (DRL). Current defenses typically rely on reward anomalies to reverse-engineer triggers and model finetuning to remove backdoors. However, complex trigger patterns undermine their robustness, and fine-tuning entails high costs, limiting practical utility. To this end, we shift defense concerns to trigger-agnostic backdoor output behaviors and propose BehaviorGuard, an online behavior-based backdoor detection and mitigation framework for DRL. Specifically, we find that regardless of attacks, backdoored policies induce consistent shifts in action distributions to ensure reliable activation, leaving detectable traces in high-quantile regions and distribution tails, even in the absence of triggers. Based on this, we design a novel metric that captures behavioral drift in action distributions to identify and suppress backdoor actions at runtime. To our knowledge, this is the first online backdoor defense that counters attacks both in single- and multi-agent DRL. Evaluated across diverse benchmarks with different backdoor attacks, BehaviorGuard consistently surpasses prior methods in both efficacy and efficiency.

Spatio-temporal forecasting is inherently bottlenecked by a phenomenon we term Spatio-Temporal Duality: the paradoxical coexistence of time-varying physical lags and instantaneous semantic synchrony. Constrained by passive coupling, existing architectures often employ indiscriminate spatial aggregation that fails to dynamically arbitrate interaction intensity based on traffic states, forcing models to rely on redundant deep stacking to approximate complex dynamics. We address this issue by introducing Time Arbitrated Spatial GNN (TAS-GNN), an efficient framework that arbitrates spatial interactions through time. Our model leverages deep temporal semantics to dynamically and proactively manage the aggregation of spatial domains, effectively decoupling physical connectivity from semantic relevance. In addition, to ensure model simplicity, we also introduce spectral decoupling via discrete wavelet Transform (DWT) to capture multiscale dependencies with minimal overhead. Experiments on three real-world datasets (PEMS04, 07, 08) show that TAS-GNN not only outperforms the current baseline model in accuracy, but also significantly improves the inference speed while reducing the number of parameters.

Spatial multi-omics technologies jointly profile diverse molecular modalities with spatial context, providing a comprehensive view of cellular heterogeneity and tissue organization. To integrate spatial multi-omics data and identify spatial domains, a wide range of unsupervised methods has been proposed. However, recent approaches rely on single-step fusion, directly aggregating heterogeneous modalities into a shared representation, making embeddings sensitive to modality imbalance and measurement noise that are inherent to spatial multi-omics data, as well as unstable optimization under weak supervision. Here, we propose ILR-SMO (Iterative Latent Refinement for Spatial Multi-Omics), a unified self-supervised framework that reformulates multimodal integration as a stability-aware refinement process over a shared latent representation. Instead of single-step plain fusion, ILR-SMO progressively integrates modality-specific information through sequential updates, allowing information to be injected in a controlled manner. This incremental refinement enables later updates to correct or compensate for earlier deviations, resulting in more stable and robust representation learning under heterogeneous and noisy modalities. ILR-SMO further incorporates reliability-aware modality gating for adaptive modulation of modality contributions, and employs a joint objective to enforce spatial coherence while preventing representation collapse. Extensive experiments on five spatial multi-omics benchmarks demonstrate that ILR-SMO consistently outperforms seven state-of-the-art methods and exhibits strong robustness across diverse settings.

Neural Architecture Search (NAS) automates the design of Vision Transformer (ViT) architectures. However, the high computational cost of training-based methods has made training-free, zero-cost proxies a key research direction. While existing proxies can estimate model potential, they fail to capture critical information transmission characteristics due to the black-box nature of neural networks. Consequently, they suffer from high computational latency and weak correlation with ground-truth performance. Although information transmission determines performance, it has not been effectively quantified or utilized. This limitation remains the core bottleneck for current zero-cost proxies. In this paper, we propose Info-NAS, a zero-cost proxy based on architectural information. This method achieves the first structured quantification of information transmission in ViT. Specifically, it evaluates architectures using three core components: global information volume, local information gradient, and global consistency. Info-NAS derives proxy scores solely from architectural parameters, requiring no forward or backward propagation. Consequently, the calculation and search processes are highly efficient. Moreover, we construct the ViT-Info-Bench dataset to facilitate correlation analysis and algorithm evaluation. Experimental results demonstrate that Info-NAS significantly reduces search overhead while achieving superior ranking accuracy. With a single evaluation requiring only 1.8 ms, Info-NAS outperforms existing methods in both efficiency and performance.

In intraday stock return forecasting, existing Transformer methods face high computational complexity and do not explicitly handle noisy predictors. We propose TransAlpha, a light-weight Transformer variant tailored for cross section data to advance end-to-end forecasting methods with three novel components (Cross-Sectional Denoising for signal purification, Temporal Attention Gating for trend weighting, and Smart Pooling to avoid information loss) along with a multi-component hybrid loss function. We validate on full A-share market stock data and four key segments (CSI 300/500/1000/2000) with 400 proprietary alpha factors of 15-minute frequencies. TransAlpha outperforms state-of-the-art baselines in predictive power and portfolio profitability, indicating tailored Transformer models have significant practical value in quantitative trade.

Hyperspectral image (HSI) clustering facilitates the unsupervised discrimination of complex surface materials but traditionally relies on the idealized assumption of centralized data availability. In real-world scenarios, however, this assumption clashes with data privacy regulations and the physical distribution of data across isolated silos. While Federated Learning offers a decentralized solution, existing frameworks in remote sensing are predominantly confined to supervised paradigms and struggle to address the heavy reliance on annotations and Non-IID distributions inherent among clients. To overcome these limitations, we propose a novel framework named Robust Federated HSI Clustering(RFHC) that enables collaborative unsupervised learning without requiring raw data exchange. Specifically, we design a dual-encoder architecture that incorporates a Federated Model Weight Augmentation strategy(FMWA), which generates consistent views through local-global network interactions to mitigate the spectral distortion introduced by traditional data augmentation. Furthermore, we develop a hybrid optimization objective that synergizes prototype relationship optimization with contrastive learning. This mechanism utilizes global prototypes to guide local training, effectively stabilizing feature learning against data heterogeneity while ensuring intra-cluster compactness. Extensive experiments on three benchmark HSI datasets demonstrate the effectiveness and superiority of the proposed method against state-of-the-art model(SOTA) federated approaches.

The learnware paradigm aims to construct a dock system that maintains a collection of learnwares, each consisting of a well-established model coupled with a specification, thereby enabling users to directly reuse existing models to solve their tasks without training models from scratch. As a core component of the paradigm, the specification sketches the model’s properties and establishes reusability between the learnware and user tasks. Existing methods have demonstrated promising performance by designing specifications that characterize the task distributions mastered by well-established models. However, in complex real-world scenarios, task distributions often overlap, thereby weakening the uniqueness of specifications and obscuring the reusability relationships between learnwares and tasks. In this paper, we design an Explicitly Distinctive Specification (EDS) that enforces specification uniqueness, improving the discriminative capability of the learnware paradigm and avoiding ambiguity caused by overlapping task distributions. This specification method explicitly leverages distributional discrepancies between specifications to strengthen the uniqueness of the model properties sketched during the submitting stage and subsequently exploits the enhanced distributional discriminability in the deploying stage to improve learnware reusability. Extensive experiments demonstrate the effectiveness and strong performance of the proposed EDS within the learnware paradigm under complex task scenarios.

Spatiotemporal forecasting of infrared thermal fields is a critical technology for the predictive maintenance of industrial electrical control equipment. However, many existing methods are designed under index-based temporal representations and global regression objectives, which are not well aligned with schedule-driven, non-stationary industrial regimes and the long-tailed spatial distribution of safety-critical hotspots. To address these issues, this paper proposes PhyRE-Net, a physical-time-anchored network architecture featuring active spatial gating. Specifically, we first design the multi-period resolution rotary positional embeddings mechanism within the backbone, which injects absolute temporal features into attention geometry to align predicted phases with non-stationary production schedules. We then devise a reinforcement learning-driven spatial gating mechanism to dynamically modulate feature responses, actively shifting the computational focus from the dominant background to sparse but critical thermal anomalies. Extensive experiments on a large-scale real-world dataset covering seven types of heterogeneous electrical equipment show that PhyRE-Net achieves consistent improvements over strong baselines across all evaluated metrics. The source code is available at https://github.com/YST-10/PhyRE-Net.

Multi-label data often contain high-dimensional features, outlier instances, and noisy labels, all of which can lead to the curse of dimensionality and decreased performance in downstream tasks. Although numerous data reduction methods have been developed, existing approaches face two major limitations: 1) existing methods typically select features, instances, or labels independently, without considering how noise or redundancy in one dimension may negatively influence the selection of others; 2) there are very few feature and instance co-selection methods that commonly assume label annotations are free of noise, which is seldom true in practice. To address these issues, we propose Evidential Multi-Label Multi-Dimensional Selection (EMMS), which jointly performs feature, instance, and label selection on multi-label data. EMMS introduces a dual projection mechanism with sparsity constraints that transforms high-dimensional data first into a latent space and then into the label space. Simultaneously, projection residuals are explicitly modeled to facilitate the identification of representative instances, enabling unified selection across features, instances, and labels. Moreover, EMMS employs evidence theory to fuse instance-level and label-level evidence, thereby enhancing the reliability of the learned labels and reducing the influence of noisy labels, which in turn promotes multi-dimensional selection. Extensive experiments demonstrate that EMMS consistently outperforms state-of-the-art methods.

Offline-to-online reinforcement learning harnesses the stability of offline pretraining and the flexibility of online fine-tuning. A key challenge lies in the non-stationary distribution shift between offline datasets and the evolving online policy. Common approaches often rely on static mixing ratios or heuristic-based replay strategies, which lack adaptability to different environments and varying training dynamics, resulting in suboptimal tradeoff between stability and asymptotic performance. In this work, we propose Reinforcement Learning with Optimized Adaptive Data-mixing (ROAD), a dynamic plug-and-play framework that automates the data replay process. We identify a fundamental objective misalignment in existing approaches. To tackle this, we formulate the data selection problem as a bi-level optimization process, interpreting the data mixing strategy as a meta-decision governing the policy performance (outer-level) during online fine-tuning, while the conventional Q-learning updates operate at the inner level. To make it tractable, we propose a practical algorithm using a multi-armed bandit mechanism. This is guided by a surrogate objective which simultaneously maintains offline priors and prevents value overestimation. Our empirical results demonstrate that this approach consistently outperforms existing data replay methods across various datasets, eliminating the need for manual, context-specific adjustments while achieving superior stability and asymptotic performance.

Masked autoencoder have demonstrated excellent performance in point cloud representation learning and received widespread attention. However, due to the unstructured nature of point clouds, existing methods struggle to effectively model both local geometric information and global topological features, while generally neglecting positional information, resulting in insufficient model representational capacity. To address these challenges, we propose a Position-Assigned Graph Masked Autoencoder (PA-GMAE) framework for point cloud representation learning. Specifically, we first explicitly transform the unstructured point clouds into a graph structure and assigns it positional information, leveraging the structured properties of graph to mine the intrinsic correlations among points. During the encoding stage, global contextual are dynamically injected into node features, endowing the model with capabilities of global awareness. In the decoding stage, collaborative work across multiple decoding tasks fully captures the global topology and positional dependencies of the point cloud, enhancing the quality of learned representations. Extensive experiments are conducted on three authoritative benchmark datasets, namely ScanObjectNN, ModelNet40, and ShapeNetPart, across typical downstream tasks including point cloud classification and part segmentation, the results demonstrate the effectiveness and superiority of the proposed framework.

Irregular Multivariate Time Series (IMTS) analysis is a challenging task as asynchronous irregular sampling disrupts intra-variable temporal consistency and cross-variable alignment. Most existing methods model multivariate correlations at either the variable or observation level in static ways. They suffer from correlation loss or cross-variable misalignment inevitably. In this paper, we propose a novel method DyH2-STGraph, Dynamic Hierarchical Heterogeneous Spatio-Temporal Graph, for IMTS forecasting. Within the graph, irregular observations are represented as nodes with spatio-temporal coordinates and observed values, connected with other neighbor observation nodes dynamically by spatio-temporal neighbor selection, while variables are treated as hyper nodes, connected with their constituent observation nodes as well as other variable nodes. The multivariate correlations are classified into the following three categories in specific manners: (1) Intra-variable coarse-to-fine correlations between variable nodes and their constituent observation nodes captured by hierarchical message propagation, (2) Inter-variable fine-grained correlations among neighboring observation nodes captured by spatio-temporal attention, and (3) Inter-variable coarse-grained correlations among variable nodes captured by attention. Extensive experiments on four benchmark datasets demonstrate that DyH2-STGraph significantly outperforms state-of-the-art methods while maintaining competitive efficiency.

Incomplete multi-view clustering (IMVC) aims to exploit complementary information from multiple views with missing observations. Recent diffusion-based approaches have shown promise for view completion; however, they often fail to capture instance-aligned global correlations across views and suffer from inefficient inference and loosely coupled optimization. In this paper, we propose IASR, an Instance-Aligned Semantic Reconstruction framework for IMVC. IASR formulates missing-view recovery as a cross-view token alignment generator, in which noisy targets, observed views, and timestep embeddings are jointly represented as tokens and interact to capture long-range cross-view dependencies throughout the denoising trajectory. To stabilize representation learning under generative noise, we further introduce a stable–active dual-encoder representation architecture with a generative-adaptive contrastive learning strategy that tightly couples view completion and clustering. Extensive experiments on eight benchmark datasets demonstrate that IASR consistently outperforms state-of-the-art IMVC methods, especially under high missing-rate settings, while achieving improved inference efficiency.

Achieving an optimal balance between exploration and exploitation remains a fundamental challenge in reinforcement learning. This work revisits the exploration-exploitation dilemma through the lens of entropy, offering a novel perspective on this enduring problem. It establishes a theoretical connection between policy's entropy and exploratory behavior, using entropy as a rational measure to quantify the exploration-exploitation trade-off. Theoretical analyses demonstrate that a modified Bellman equation, augmented with a novelty-seeking term, ensures appropriate entropy adjustment and guarantees its globally monotonic decay. The derived policy optimization process inherently accommodates all three regimes of the exploration-exploitation spectrum, enabling a principled transition from exploration to exploitation. Building on these theoretical insights, this work introduces AdaZero, an adaptive deep architecture that dynamically balances exploration and exploitation. Extensive empirical evaluations highlight AdaZero's robust performance and validate the theory's feasibility.

Many scientific and engineering applications require estimating unknown parameters from experimentally observable data -- an inverse problem that is inherently challenging due to nonlinearity, noise, and ill-posedness. In this paper, we propose an active diffusion-based inverse problem solver. A diffusion model is trained to learn the mapping between the parameter space and the observable space. By iteratively detecting and correcting model misspecification through posterior uncertainty, the method discovers and learns the correct region of parameter space, even when initial training bounds exclude the true parameters. This provides a principled, Bayesian justification for adaptive domain augmentation and ensures robust inference for inverse problems under incomplete prior knowledge. We demonstrate the effectiveness of our inverse solver for a toy inverse problem with infinite solutions, and for the parameterization of the quantum correlation functions to event observables in a Quantum Chromodynamics analysis of nucleon structure.

Vision-Language-Action (VLA) models bridge multimodal reasoning with physical control, but adapting them to new tasks with scarce demonstrations remains unreliable. While fine-tuned VLA policies often produce semantically plausible trajectories, failures often arise from unresolved geometric ambiguities, where near-miss actions lead to divergent execution outcomes under limited supervision. We study few-shot VLA adaptation from a generation-selection perspective and propose a novel framework, VGAS (Value-Guided Action-chunk Selection). It performs inference-time best-of-N selection to identify action chunks that are both semantically faithful and geometrically precise. Specifically, VGAS employs a finetuned VLA as a high-recall proposal generator and introduces Q-Chunk-Former, a geometrically grounded Transformer critic to resolve fine-grained geometric ambiguities. In addition, we propose Explicit Geometric Regularization (EGR), which shapes a discriminative value landscape to preserve action ranking resolution among near-miss candidates while mitigating value instability under scarce supervision. Experiments and theoretical analysis demonstrate that VGAS consistently improves success rates and robustness under limited demonstrations and distribution shifts. Our code is available at https://github.com/Jyugo-15/VGAS

The disaggregated Prefill-Decode (PD) architecture has emerged as a prominent paradigm for efficient Large Language Model inference serving. However, resource management remains a critical challenge, particularly under the dual burstiness of real-world scenarios—characterized by volatile fluctuations in both request arrival rates and Prompt-to-Response ratios. Existing rule-based heuristics often fail to accurately identify system bottlenecks, leading to severe resource misallocation and Service Level Objective (SLO) violations. To address this, we propose a Deep Reinforcement Learning-based auto-scaling framework tailored for the PD architecture. By modeling the resource allocation problem as a Markov Decision Process, our framework enables the agent to capture non-linear load dynamics, thereby achieving decoupled and precise scaling for prefill and decode pools. Furthermore, to mitigate Head-of-Line blocking caused by scaling latency, we design an immediate rescheduling mechanism that migrates queued tasks to newly ready nodes in real-time. Experimental results driven by Azure bursty load traces demonstrate that our framework significantly reduces computational costs by 25.2% and 28.3% compared to the Static configuration and HeteroScale, respectively, while strictly adhering to SLOs.

Accurate prediction of prerequisite relations among concepts is important for course planning and intelligent tutoring systems. Existing text-based methods are frequently contaminated by noise such as redundant phrasing, ambiguous sentences, and domain-specific colloquialisms. Moreover, previous methods based on graph structures may have failed to capture the complex interactions between concepts and documents, an information gap exists between the two approaches. Therefore, we propose an LLM Enhancement and Concept-Document Interactive Modeling for Prerequisite Relation Prediction (LECDPR) model. Concretely, LECDPR elicits an LLM via two-stage prompting to distill concept descriptions and contextual evidence from documents, resulting in semantic space embeddings of concepts. Then, the complex interactive relations between concepts and documents are modeled through three propagation mechanisms to produce relational space embeddings, aligns the two embeddings through random masking and contrastive learning, and adaptively fuses the resulting embeddings. Finally, the fused concept representations are used for more accurate prerequisite relation prediction. Extensive experiments on the public UCD, Lecture Bank, and MOOC datasets show that LECDPR outperforms ten baselines on ACC, AUC, and F1. Our code is available at https://github.com/lecdpr/LECDPR.