IJCAI 2026 Papers

Exploiting unlabeled data through semi-supervised learning (SSL) or leveraging pre-trained models via fine-tuning are two prevailing paradigms for addressing label-scarce scenarios. Recently, growing attention has been given to combining fine-tuning of pre-trained vision-language models (VLMs) with SSL, forming the emerging paradigm of semi-supervised fine-tuning. However, existing methods often suffer from model bias and hyperparameter sensitivity, due to reliance on prediction consistency or pre-defined confidence thresholds. To address these limitations, we propose a simple yet effective plug-and-play methodology named Bi-Consistency-Guided Self-Training (Bi-CoG), which assigns high-quality and low-bias pseudo-labels, by simultaneously exploiting inter-model and intra-model consistency, along with an error-aware dynamic pseudo-label assignment strategy. Both theoretical analysis and extensive experiments over 14 datasets demonstrate the effectiveness of Bi-CoG, which consistently and significantly improves the performance of existing methods.

Multi-view clustering aims to discover shared semantic structures from multiple complementary data views to improve clustering performance. However, most existing methods rely on a single clustering representation for each semantic cluster, which limits the ability to model complex cluster structures and diverse patterns in real-world data. To address this limitation, we propose a Mixture of Clustering Experts with Dual Consistency (MoCE-DC). MoCE-DC introduces a set of learnable clustering experts that characterize the same semantic cluster from multiple perspectives within a unified prototype space. A gating mechanism is employed to adaptively select experts for each sample. MoCE-DC decomposes complex semantic clusters into multiple collaborative sub-structures, significantly enhances the ability to model intricate intra-cluster diversity. To further ensure cross-view semantic consistency, we propose a dual-level alignment mechanism that enforces prediction consistency across views while guiding clustering assignments toward a more discriminative direction. In addition, a gating balance regularization strategy is introduced to mitigate expert collapse and promote balanced expert utilization. Extensive experiments on multiple public multi-view clustering benchmarks demonstrate that MoCE-DC outperforms state-of-the-art methods.

Standard unsupervised multi-view feature selection (UMFS) methods for large datasets exhibit limitations in modeling the competition between cross-view alignment and intra-view diversity, resulting in suboptimal solutions and expensive computational costs. This challenge is exacerbated by the diverse signals inherent in complex samples, which tend to mask shared patterns, thus complicating the pursuit of an optimal trade-off. In this work, we propose a Progressive Adversarial Feature Selection (ProAd-FS) framework for large-scale multi-view learning, which formulates this static trade-off objective as a dynamic competitive process. To be specific, ProAd-FS develops an adversarial decoupled architecture that assigns these competing objectives to distinct inter-view and intra-view games, guided by a robust gated curriculum to prioritize the learning of underlying structures. The entire framework exhibits linear computational complexity in both sample size and feature dimension, and embeds structural sparse regularization for end-to-end optimization. Extensive experimental results show that ProAd-FS achieves state-of-the-art performance while being highly scalable as well, providing an effective solution for large-scale UMFS tasks.

Existing spatial domain identification methods primarily use graph neural networks to model spatial and transcriptional relationships. However, their performance is highly sensitive to noisy affinity graphs. Moreover, graph autoencoders tend to over-constrain latent representations, which limits their ability to capture global variability in spatial multi-omics data. To overcome these issues, we propose a dual-flow latent refinement framework that simultaneously integrates structure-aware and structure-free transformations. Specifically, a graph normalizing flow is employed to enforce relational consistency, while a parallel vanilla normalizing flow preserves global distributional flexibility. Features learned by the two flows are then adaptively fused to obtain a robust and unified latent representations. In addition, we introduce a grouping belief-based affinity refinement strategy to suppress unreliable connections and strengthen confident neighborhood relationships, which provides a more stable structural prior for representation learning. Extensive experiments on multiple spatial multi-omics datasets show that the proposed method consistently outperforms state-of-the-art approaches and achieves more accurate and robust spatial domain identification. The supplementary material is publicly available at https://github.com/LiangSDNULab/GBFlow.

Probabilistic time series forecasting seeks to quantify the uncertainty of future observations. While recent works introduce latent variables to alleviate the spurious dependencies caused by hidden confounders, thereby reducing overly wide confidence intervals, simply incorporating latent factors is not sufficient. When the underlying latent dynamics evolve at multiple temporal scales, existing methods may entangle temporally coarse and fine latent dynamics, which introduces spurious latent transitions and in turn amplifies predictive uncertainty. Therefore, disentangling temporally coarse and fine latent dynamics is essential for achieving sharper and more reliable probabilistic forecasting results. Building on this insight, we propose COFE (COarse And FinE latent dynamics disentanglement), a variational autoencoder–based framework that models the temporal distribution by disentangling and modeling the rapidly changing and slowly varying latent dynamics simultaneously. In particular, the proposed COFE harnesses the independence of estimated noises between adjacent latent states as well as a sparsity constraint on estimated noise to disentangle latent dynamics across different scales effectively. More specifically, we show that the multi-scale latent dynamics are disentangled with rigorous theoretical guarantees. Extensive experiments on 15 benchmark datasets, compared against 12 state-of-the-art baselines, demonstrate that COFE achieves superior uncertainty quantification, validating its effectiveness in a wide range of real-world scenarios. Code is available at https://github.com/polars8948/COFE

Zero-shot transfer for offline reinforcement learning involves generalizing to a wide range of tasks that are often risk-sensitive, without new interaction. We show here that although Successor Features (SFs) provide a principled framework for transfer through abstracting dynamics from rewards, their typical form is restricted to expected cumulative returns and is thus indifferent to variance in returns. We contend that such a mean-centric approach introduces bias by conflating risky and safe trajectories with identical expected outcomes, as it ignores outcome dispersion critical for risk-aware evaluation. Additionally, we note that distributional methods can be employed to model outcome distributions, but frequently do so at a high computational cost due to the need for discretization or elaborate sampling. To overcome this problem, we introduce a form of successor representation that incorporates an uncertainty envelope over states: Gaussian Distributional Successor Features (GDSF). Rather than inferring an implicit generative process, GDSF learns to approximate the distribution of cumulative outcomes with a Gaussian through recursive moment-matching. We present this framework for efficient representation of aleatoric uncertainty, without resorting to computationally expensive existing distributional methods. We also introduce Latent Outcome Optimization, an inference procedure which finds high-reward outcomes in the data support and drives a goal-conditioned policy. This dual-mode design allows the agent to transfer to non-linear objectives in a zero-shot way while preserving closed-form efficiency for linear tasks. Experiments on the D4RL benchmark show that GDSF yields better performance in risk-sensitive tasks while achieving at least competitive results in standard linear regimes.

Molecular property prediction is a fundamental task with wide-ranging applications in chemistry, materials science, and drug discovery. Beyond predictive accuracy, the reliability of model predictions, particularly their uncertainty awareness, is critical for high-stakes downstream applications. However, most existing methods predominantly optimize performance metrics while overlooking prediction reliability, often producing results that are both inaccurate and overconfident. To address this challenge, we propose an uncertainty-guided framework that integrates adaptive local adversarial perturbation with evidential deep learning for reliable molecular property prediction. The proposed method identifies sensitive molecular regions through gradient-based sensitivity analysis and applies locally adaptive perturbations whose magnitudes are dynamically modulated by predictive uncertainty. This approach explicitly challenges the model to preserve predictive stability under structure-aware perturbations, thereby enhancing its sensitivity to critical molecular substructures. Furthermore, we introduce an evidential calibration strategy that aligns uncertainty with the prediction discrepancy between clean and adversarial molecular views, enabling the learned evidence to faithfully reflect prediction stability. Extensive experiments on benchmark molecular property prediction datasets demonstrate that our approach achieves state-of-the-art predictive performance while providing more reliable uncertainty estimates than widely used uncertainty quantification methods. These results highlight the potential of the proposed framework as a reliable tool for risk-sensitive virtual screening and molecular decision-making. Our code is available at https://github.com/Ubehind/ugap.

Similarity-graph-based multi-view clustering is effective in capturing non-Gaussian cluster structures, yet it suffers from an inherent conflict between scalability and geometry-consistent cluster assignment. In particular, cluster labels are typically inferred via secondary clustering rather than being learned directly from the underlying data geometry. Although anchor-based extensions alleviate the computational burden, they introduce the challenge of anchor-number selection. To address these limitations, we propose Multi-view Clustering via Manifold Decomposition (MvMD), which directly infers cluster labels from multi-view data without explicit similarity graph construction or anchor selection. MvMD reformulates multi-view clustering as a unified multi-view regression problem, where cluster labels are optimized as model variables. To improve robustness against missing small clusters, a global balance regularization is incorporated. Meanwhile, local geometric consistency and the low-rank structure of cluster assignments are jointly enforced through a symmetric matrix factorization scheme, which is efficiently realized via a truncated SVD-based low-rank approximation. Extensive experiments on benchmark datasets validate the effectiveness and efficiency of the proposed MvMD. The code is available at https://github.com/Vince-Doit/MvMD.

Spatially resolved transcriptomics integrates gene expression with spatial coordinates to decode tissue microenvironments. Existing methods predominantly utilize graph structures to model relationships between spots. However, their performance is bottlenecked by the reliability of gene feature graph, facing the following hurdles: (1) ubiquitous housekeeping genes cause high-expression spots to densely connect with heterogeneous spots, leading to a skewed graph structure; (2) information reduction during Highly Variable Gene (HVG) selection results in the loss of intrinsic local structures. To address these challenges, we propose a Spot-Adaptive Structural Rectification method, called SASR. Specifically, SASR employs a hyperspherical expansion constraint that projects gene expression profiles onto a unit hypersphere to maximize angular distances, effectively separating spots falsely clustered by high total counts. Simultaneously, a topological consistency constraint repairs structural fractures caused by HVG selection via aligning latent embeddings with the local structures of the raw full-gene space. The complementary synergy balances angular discriminability with topological fidelity for accurate clustering. Experiments demonstrate that SASR effectively corrects structural biases and surpasses state-of-the-art methods in spatial clustering.

As graph repositories grow in scale and diversity, training Graph Neural Networks (GNNs) becomes computationally demanding. However, existing graph condensation methods often fail to retain the intrinsic structural patterns of the original graphs, which are essential in graph-based learning. Therefore, these methods suffer from limited performance and poor generalization in downstream tasks due to the loss of structural information. To address this, we propose Similarity-guided Structural Matching Learning for Graph Dataset Condensation (SSGDC), which efficiently reduces repository size while maintaining both task performance and structural information. Our approach introduces a similarity-based graph selector to identify high-quality subsets for condensation. The condensed graphs are optimized using a dual-objective loss that combines gradient-matching for task alignment with a metric-learning loss for structural preservation within the selected subset. This ensures that the condensed dataset retains both task-relevant information and the essential relational topology that supports GNN training and enhances generalization. Experiments demonstrate that our method achieves higher accuracy and better structure retention across varying condensation ratios, highlighting the critical role of structural preservation in graph dataset condensation.

In many real-world scenarios, agents operate across heterogeneous environments where the underlying dynamics and data-generating processes vary. Standard reinforcement learning and imitation learning methods often fail in such settings as they typically assume stationarity and learn policies that overfit to environment-specific correlations. A key challenge is the presence of spurious correlations as observed states often contain both causal and non-causal features, with the latter introducing environment-specific biases that undermine generalization. To address this problem, we propose DropConnect-based Causal Imitation Learning (DCIL), a novel offline imitation learning framework designed to identify and exploit stable causal mechanisms across diverse environments. DCIL introduces a gradient alignment constraint that encourages the policy to align with causal structures shared across training environments. To further mitigate overfitting to spurious correlations, DCIL involves DropConnect-based regularization, injecting stochastic perturbations into network weights to simulate parameter uncertainty and reduce reliance on unstable features. We evaluate DCIL on synthetic benchmarks derived from OpenAI Gym control tasks, where non-causal features exhibiting spurious correlations are explicitly injected to simulate environmental heterogeneity. Experimental results show that DCIL outperforms state-of-the-art imitation learning baselines, achieving superior generalization to unseen environments. These findings highlight the importance of incorporating causal reasoning and structured regularization into policy learning for robust performance under environment shift.

Open cross-domain federated graph learning facilitates collaborative learning among clients from distinct graph domains while preserving privacy. However, severe structure and feature heterogeneity in open scenarios exacerbates the multiplicative amplification of structural and feature noises within graph neural network encoders, leading to significant negative transfer. Existing methods addressing partial heterogeneity are insufficient. We theoretically reveal this amplification mechanism and propose a novel learning method FedFINFO, which leverages full-informativeness shared knowledge. Specifically, we construct an aligned and general feature space integrating node semantic and structural roles. We proposed a counterfactual multi-view framework to explicitly learn a structure masker for extracting consensus subgraphs. These jointly suppress the mutual amplification of noises at the source. We also propose a history-aware adaptive multi-channel aggregation mechanism to enable dynamic and personalized sharing driven by channel similarity while preventing collaborative oscillation. FedFINFO effectively reconciles common knowledge sharing with personalized knowledge retention, significantly mitigating negative transfer. Extensive experiments under cross-dataset and cross-domain settings demonstrate the superiority of our method.

Class-Incremental Learning (CIL) aims to enable models to sequentially learn new tasks while retaining knowledge from previous ones. Recently, merging-based pre-trained CIL methods have gained significant attention due to their competitive performance and high inference efficiency. However, most existing approaches decouple training from merging, neglecting the compatibility among task-specific adapters. This incompatibility introduces severe conflicts during integration, resulting in catastrophic forgetting and notable performance degradation. To address this limitation, we propose MOBO, a Merging-Oriented Bi-Level Optimization framework that synergizes the optimization of the current task model with the performance of the final merged model. At the upper level, we introduce a global loss that anticipates the merged model's behavior, guiding the current task parameters toward a solution space that facilitates effective merging. At the lower level, we employ a dynamic weighted merging strategy to optimize merging coefficients and update the merged model. By alternately optimizing the task-specific and merged models, MOBO effectively mitigates the performance loss caused by incompatible adapter integration. Comprehensive experiments on CIFAR-100, CUB-200, ImageNet-R, ImageNet-A, and VTAB demonstrate the superiority of our approach, highlighting its robustness and scalability, particularly on long task sequences.

Graph Masked AutoEncoders (GMAEs) have achieved notable success in diverse downstream tasks. However, their superior performance usually relies on reliable and unperturbed input graphs. In real-world scenarios, graphs are vulnerable to adversarial attacks (e.g., perturbations on node features or topology). As a result, existing GMAEs tend to overfit these perturbed inputs, leading to poor robustness. Motivated by this, we propose an adversarial masked graph modeling for robust graph autoencoders, named ArmorGAE. Specifically, we first design a dual-consistency graph augmentation method that enriches the graph topology to mitigate structural vulnerabilities. Theoretically, we establish a connection between ArmorGAE and contrastive learning. It demonstrates that our augmentation strategy in GMAEs is equivalent to effectively expanding the set of high-quality positive samples in contrastive learning, providing richer and more reliable supervisory signals. We then present an adversarial masked autoencoder to mitigate model overfitting to the perturbed graph. The generator and discriminator are jointly trained under an adversarial paradigm, where the former aims to recover the underlying true data distribution, while the latter distinguishes reconstructed and real edges. Experimental results on five datasets demonstrate that ArmorGAE outperforms state-of-the-art methods in terms of classification accuracy on clean graphs and adversarial robustness under 16 distinct adversarial scenarios. The codes are publicly available at https://github.com/ZZY-GraphMiningLab/ArmorGAE.

The Spherical Sliced Wasserstein (SSW) distance has emerged as a fundamental tool for comparing probability measures on the hypersphere, with broad applications spanning geology, medical imaging, computer vision, and representation learning. However, its reliance on uniformly sampling a large number of projection directions from the unit hypersphere can lead to suboptimal performance, as many directions are weakly informative and fail to capture salient differences between distributions. We address this limitation through Fusion Stereographic Spherical Sliced Wasserstein (FS3W), a data-adaptive divergence that incorporates distribution-dependent information into projection selection. FS3W uses a neural slicing fusion mechanism to refine prior directions, steering them toward projections aligned with discriminative geometric structures in the data. This adaptive approach produces more expressive and accurate characterizations of distributional discrepancies on the hypersphere. We evaluate FS3W across five diverse tasks, including gradient flows, Earth density estimation, and self-supervised representation learning. Empirical results consistently demonstrate that FS3W outperforms existing spherical optimal transport–based methods.

As a prominent paradigm for large-scale unsupervised learning, anchor-based multi-view clustering aims to reveal the latent structures across heterogeneous data representations with high efficiency. Despite achieving some progress, existing methods typically suffer from the following two limitations. Firstly, they rely on pre-constructed anchors or rigid constraints (e.g., orthogonality), whereas the intrinsic topological correlations among anchors are completely discarded. Secondly, most existing methods either perform clustering directly on the bipartite graph or treat different views equally, thereby failing to capture the global structural information. To this end, the Dual-tOpology learning with adapTive Anchors (DOTA) is proposed, which not only learns the sample-anchor relationship (i.e., bipartite graph) but also preserves the topology structure among anchors, significantly enhancing the discriminability of learned representation while preserving the underlying data manifold. By integrating an adaptive view-weighting strategy to balance view contributions, DOTA derives discriminative global sample embeddings by propagating spectral information through the bipartite graph. Extensive experiments on benchmark datasets demonstrate the superiority of DOTA.

Recent personalized federated learning research focuses on heterogeneous models across clients. However, existing methods often rely on external data, model decoupling, and partial learning, which makes them sensitive to settings. In contrast, we revisit hypernetworks and leverage their strong generalization ability to propose the first practical method for personalized federated learning. We first propose a model heterogeneous personalized federated learning framework based on hypernetworks, MH-pFedHN, which quantifies clients with different architectures using customized embedding vectors and then generates client-specific model parameters through a server-side hypernetwork. Besides the shared feature extractor, our hypernetwork consists of multiple heads, where clients with similar numbers of parameters are assigned the same number of customized embedding vectors and consequently share the same head. This design enables knowledge sharing across different architectures and reduces the computation of parameter generation. To further enhance the hypernetwork’s learning and generalization, we propose MH-pFedHNGD, which introduces a lightweight yet effective plug-in global model. Our framework requires no external data and does not disclose client model architectures, thereby effectively ensuring security and demonstrating great potential. Experiments across various models and tasks demonstrate that our approach outperforms standard baselines and exhibits strong generalization performance. Our code is available at https://github.com/DangDang1895/MH-pFL

Continual test-time adaptation (CTTA) seeks to adapt a source-pretrained model to continuously shifting target distributions. Recent advances in this field either minimize prediction entropy on streaming target samples, which is efficient but prone to error accumulation, or rely on teacher–student pseudo-labeling, which provides more stable predictions at the cost of high computational overhead. To alleviate these issues, in this paper, we propose Dual Alignment (D-Align), a method that jointly optimizes correlation alignment and masked consistency alignment for stable and efficient online adaptation. Specifically, correlation alignment constructs a reliable pseudo-source domain by selecting resilient target samples, maintaining their diversity via a dynamic memory bank, and aligning nonlinear feature relations through image-level visual domain prompt–driven alignment, enabling effective feature alignment without costly backpropagation. Masked consistency alignment enforces prediction consistency across multiple masked views of each target sample, promoting robust and discriminative target representations. Using multiple standard benchmarks, the experiments have verified the effectiveness of D-Align under dynamic domain shifts, in comparison with state-of-the-art methods.

Predicting the response of anticancer drugs for specific cancer cell lines remains an important challenge in cancer treatment, largely due to the diversity among cell lines. Existing machine learning methods usually perform well on cell lines seen during training, but their performance often drops when applied to new cell lines. Many approaches rely heavily on cell-line specific features, which limits their ability to generalize. At the same time, methods that focus mainly on generalization often fail to capture differences in drug response across distinct cellular environments. Some studies improve prediction accuracy by introducing additional and more detailed cell line features. However, these features are difficult to collect and are frequently missing, especially for newly studied cell lines. To address these issues, we propose a generalizable bias-disentangled framework for drug response prediction. Our method separates drug-related patterns that are shared across cell lines from representations that reflect cell-line specific environments. We further isolate real but non-generalizable biases that arise from noise or unobserved factors. This design helps preserve accurate predictions for specific cell lines while improving performance on unseen ones. In addition, the bias-related branch supports counterfactual consistency learning when some modalities are missing, allowing the model to handle incomplete data. Experimental results show that GBD-DRP achieves state-of-the-art performance, and additional tests on unseen cell lines and missing-modality settings indicate good generalization in practice.