IJCAI 2026 Papers

Federated Learning (FL) enables distributed learning across multiple clients without sharing raw data. When statistical heterogeneity across clients is severe, Clustered Federated Learning (CFL) can improve performance by grouping similar clients and training cluster-wise models. However, most CFL approaches rely on multiple communication rounds for cluster estimation and model updates, which limits their practicality under tight constraints on communication rounds. We propose Data Collaboration-based Clustered Federated Learning (DC-CFL), a single-round framework that completes both client clustering and cluster-wise learning, using only the information shared in DC analysis. DC-CFL quantifies inter-client similarity via total variation distance between label distributions, estimates clusters using hierarchical clustering, and performs cluster-wise learning via DC analysis. Experiments on multiple open datasets under representative non-IID conditions show that DC-CFL achieves accuracy comparable to multi-round baselines while requiring only one communication round. These results indicate that DC-CFL is a practical alternative for collaborative AI model development when multiple communication rounds are impractical. Our source code is publicly available at https://github.com/soutasuga/DC-CFL.

Gradient boosting remains a strong and widely used method for tabular data learning, but its performance often degrades when training labels are noisy. This behavior is largely related to the way boosting algorithms emphasize samples with large gradients, without explicitly accounting for whether such errors originate from informative hard cases or from unreliable labels. We address this issue by reconsidering how sample reliability is evaluated during boosting. Instead of relying on instantaneous error, we examine the evolution of each sample’s residuals across iterations. Based on this insight, we propose Information-Theoretic Trust Boosting (ITBoost), which uses the Minimum Description Length principle to measure the complexity of residual trajectories. Samples whose residual patterns fluctuate in an irregular manner are treated as less trustworthy and are down-weighted during learning. Theoretically, we derive a tighter generalization bound for ITBoost under label noise. Empirical results on various tabular benchmarks indicate that ITBoost provides improved robustness in noisy environments over leading boosting and deep tabular models, while retaining best average performance on clean data.

Survival models on networks describe the time-evolving dynamics of complex systems, but existing frameworks suffer from two key limitations: First, many past approaches implicitly assume independent censoring, which can introduce substantial bias. Second, point-process-based models often emphasize only sending behavior and overlook shared latent structure between interacting nodes. Recent work has incorporated copulas into deep survival models to mitigate censoring bias, but these methods do not capture topological role structure. Another line of research combines survival analysis with the Mixed-Membership Stochastic Blockmodel (MMSB) to model heterogeneity via role-specific hazard rates; however, it typically assumes structural independence and therefore misses important correlations between roles. To bridge these gaps, we apply an improved Copula–MMSB mechanism and propose the first Survival Fully-Collapsed Copula Mixed Membership Blockmodel. Our framework jointly models time-to-event dynamics and complex dependencies among latent roles. On the Enron benchmark, it surpasses state-of-the-art survival blockmodels in both predictive likelihood and structural interpretability. Additionally, the model admits an exact closed-form discrete distribution over role pairs, which enables a fully collapsed Gibbs sampler and avoids the slow mixing and poor convergence common in non-conjugate inference. In simulations, this yields a 100× improvement in sampling efficiency over semi-collapsed baselines.

Data selection is a key component of efficient instruction tuning for large language models, as recent work has shown that data quality often matters more than data quantity. Accordingly, prior studies have introduced various multi-dimensional heuristics to evaluate and filter instruction data. However,most existing methods rely on static task-agnostic and model-agnostic weighting schemes, which overlook the varying requirements of specific downstream tasks and the differing pre-existing capabilities of models. In this paper, we propose a framework for learning multi-indicator weights that jointly adapts data selection to both the downstream task and the specific model. Our method identifies optimal weight configurations without full-scale fine-tuning by utilizing in-context learning (ICL) signals on compact tiny-validation sets. These signals serve as efficient performance proxies that ensure high-fidelity evaluation at minimal computational cost. Experiments across multiple benchmarks and model families, including Mistral, Qwen, and Llama, show that the approach achieves performance comparable to or exceeding full-dataset tuning while using only 30% of the training samples on GSM8K. Furthermore, our analysis reveals a trade-off between semantic diversity and logical complexity in reasoning tasks, highlighting the necessity of joint task-model adaptation.

Identifying meaningful latent representations within diffusion models remains a challenging problem for causal approaches. We propose Causal Manifold Transport Diffusion Model (CMT-Diff), a framework that operationalizes causal actions as geometric transformations. By adopting the perspective of backtracking counterfactuals, we formulate the generative process as a composite diffeomorphism that couples the Probability Flow ODE with a Continuous Normalizing Flow. This mapping constructs an exogenous manifold where causal factors align with coordinate variations. Within this geometry, we derive Causal Manifold Transport (CMT) to realize interventions as linear vector translations along factor-aligned directions. We establish theoretical identifiability guarantees and demonstrate that our approach facilitates controllable generation by capturing the underlying causal manifold.

For minimally rigid graphs, the same edge-length data can admit multiple realizations (up to translations and rotations). Finding graphs with exceptionally many realizations is an extremal problem in rigidity theory, but exhaustive search quickly becomes infeasible due to the super-exponential growth of the number of candidate graphs and the high cost of realization-count evaluation. We propose a reinforcement-learning approach that constructs minimally rigid graphs via 0- and 1-extensions, also known as Henneberg moves. We optimize realization-count invariants using the Deep Cross-Entropy Method with a policy parameterized by a Graph Isomorphism Network encoder and a permutation-equivariant extension-level action head. Empirically, our method matches the known optima for planar realization counts and improves the best known bounds for spherical realization counts, yielding new record graphs.

Fourier neural operators (FNOs) are effective and efficient surrogates for approximating solutions of PDEs and generalize across discretizations. However, owing to the reliance on frequency truncation to maintain learning efficiency of FNOs, empirical studies suggest that FNOs exhibit spectral bias toward low-frequency information, which may hinder the learning capability especially for certain PDEs with strong high-frequency oscillations. To address this limitation, we propose SirenFNO, a novel framework that leverages sinusoidal representation networks (SIRENs) to learn implicit neural representations and performs mode-wise kernel parameterization. Our SIREN parameterization learns a full-grid spectrum with a constant and discretization-independent parameter count, thereby eliminating the need for frequency truncation. We further extend SirenFNO with functional tensor decompositions to enhance parameter and learning efficiency. Empirical results show that our SirenFNO consistently outperforms FNO with approximately 4 to 15 times parameter reductions with preserved discretization invariance, and our functional decomposition variants obtain performance improvements with a maximum of 73 times fewer parameters across multiple PDE benchmarks.

Single-cell perturbation screens enable systematic discovery of gene regulatory mechanisms, yet the exponential expansion of perturbation space makes comprehensive experimentation impractical. Although in silico predictors have been increasingly proposed to address this challenge, most existing methods either assume static regulatory priors or learn unconstrained data-driven representations. As a result, they often suffer from poor adaptation to state-dependent shifts and weak out-of-distribution generalization. In response, we propose URFPert, an unrolled regulatory flow network tailored for single-cell perturbation prediction. Specifically, URFPert is composed of three fundamental modules: (i) Regulatory Structure Unrolling module, which employs an iterative unrolling optimization algorithm to infer state-dependent gene regulatory networks (GRNs); (ii) Heterogeneous Graph Flow Network, which incorporates the inferred GRNs into a conditional velocity field to yield state-consistent and perturbation-specific trajectories; (iii)Discriminative Optimization Strategy, which combines Mean Squared Error~(MSE) with Maximal Coding Rate Reduction~(MCR²) to promote discriminative perturbation responses and mitigate training confounders. Comprehensive evaluations on five single-cell perturbation datasets demonstrate that URFPert outperforms state-of-the-art methods in unseen perturbation regimes, providing a powerful tool for interpreting regulatory mechanisms. The project code is public via https://github.com/Jiniretttt/URFPert.

Federated learning (FL) has emerged as a promising paradigm for adapting large language models (LLMs) to distributed data. To mitigate the high communication and memory overhead, parameter efficient techniques such as Low Rank Adaptation (LoRA) are widely adopted. However, practical deployments often exhibit rank and data heterogeneity, making direct aggregation of LoRA updates biased and unstable. Existing approaches enforce a unified rank or align heterogeneous updates into a single shared subspace, which undermines personalization and accuracy. Moreover, under differential privacy (DP), adding noise to such mixed updates could perturb client specific directions that should remain local, resulting in utility loss. To address these issues, we propose Selective Decoupled Federated LoRA (SDFLoRA), a structure aware LoRA framework that decouples each client update into a shared component for updating and a private component that preserves client specific semantics. Subspace alignment and aggregation are applied selectively to the shared module, while private modules remain local. We further design low rank re-compression to maintain a fixed rank budget for the shared module. This structure supports privacy aware optimization by injecting DP noise exclusively into the shared module. Experiments on multiple benchmarks demonstrate that SDFLoRA beats federated LoRA baselines and exhibits strong robustness under privacy constraints.

Privacy concerns in distributed learning often lead clients to return intentionally altered gradient information. We consider the problem of learning convex and L-smooth functions under adversarial gradient perturbation, where a client's gradient reply to a server query can deviate arbitrarily from the true gradient subject to a distance bound. Our study focuses on two fundamental questions: (i) what is the smallest achievable sub-optimality gap (i.e., excess error in optimization) under such responses, and (ii) how many queries are sufficient to guarantee a given sub-optimality gap? We establish tight feasibility thresholds on the sub-optimality gap and provide algorithms that achieve these thresholds with provable query complexity guarantees.

Chemical Language Models (CLMs) pre-trained on large-scale molecular data are widely used for molecular property prediction. However, the common belief that increasing training resources, such as model size, dataset size, and training compute, improves both pre-training loss and downstream task performance has not been systematically validated in the chemical domain. In this work, we evaluate this assumption by pre-training CLMs while scaling training resources and measuring transfer performance across diverse molecular property prediction (MPP) tasks. We find that while pre-training loss consistently decreases with increased training resources, downstream task performance shows limited improvement. Moreover, alternative metrics based on the Hessian or loss landscape also fail to estimate downstream performance in CLMs. We further identify conditions under which downstream performance saturates or degrades despite continued improvements in pre-training metrics, and analyze the underlying task-dependent failure modes through parameter space visualizations. These results expose a gap between pre-training-based evaluation and downstream performance, and emphasize the need for model selection and evaluation strategies that explicitly account for downstream task characteristics. The code is available at https://github.com/sagawatatsuya/MolScaleTransfer.

Understanding the training dynamics of deep neural networks remains a major open problem, with physics-inspired approaches offering promising insights. Building on this perspective, we develop a thermodynamic framework to describe the stationary distributions of stochastic gradient descent (SGD) with weight decay for scale-invariant neural networks, a setting that both reflects practical architectures with normalization layers and permits theoretical analysis. We establish analogies between training hyperparameters (e.g., learning rate, weight decay) and thermodynamic variables such as temperature, pressure, and volume. Starting with a simplified isotropic noise model, we uncover a close correspondence between SGD dynamics and ideal gas behavior, validated through theory and simulation. Extending to training of neural networks, we show that key predictions of the framework, including the behavior of stationary entropy, align closely with experimental observations. This framework provides a principled foundation for interpreting training dynamics and may guide future work on hyperparameter tuning and the design of learning rate schedulers.

Although Large Language Models (LLMs) excel at code generation, recent research reveals that they exhibit an insufficient grasp of core programming concepts, such as data flow and control flow. This limitation undermines their robustness when encountering variations in these concepts in practice; however, effective solutions that explicitly target this gap remain limited. To address this challenge, we propose ProCURE, a concept-aware consistency learning framework designed to enhance LLMs’ understanding of programming concepts. Specifically, ProCURE first performs automated concept-oriented code augmentation to construct a concept-aligned dataset covering representative programming concepts. It then conducts concept-aware fine-tuning, encouraging the model to capture fine-grained concept variations and learn appropriate generation behaviors under such variations via a novel concept-sensitive consistency loss. To quantify programming concept understanding, we introduce the Concept Consistency Score (CCScore), defined as the proportion of correct generations preserved under concept variations. A higher CCScore indicates a more profound understanding of programming concepts. We evaluate ProCURE on four open-source LLMs across three widely used code generation benchmarks. Experimental results show that ProCURE improves CCScore by an average of 17.9 points, demonstrating its effectiveness in addressing the programming concept understanding gap.

In Partial Label Learning (PLL), each instance is associated with a candidate label set, with exactly one label being true. While most studies implicitly assume balanced class distributions, real-world data often exhibit severe class imbalance distributions, leading to the Long-Tailed Partial Label Learning (LT-PLL) problem. In response, prevailing strategies typically manifest the label space to retrieve more tail samples for improved pseudo-label accuracy. However, hard-to-distinguish tail samples can significantly hinder representation learning and, in turn, affect the quality of pseudo-labels. To address this, we propose a novel Distribution-aware Dual-branch Contrastive Learning framework, DEACON, that decouples representation of head and tail labels via a dual-branch mutual-teaching design, enabling disambiguation across different shot-level groups with tailored representations. DEACON comprises two modules: (i)-a instance-balanced branch trained with a vanilla contrastive objective; (ii)-a label-balanced branch trained with a novel reweighted loss that adjusts the contrastive strength through the conditioned von-Mises Fisher density. The two branches mutually teach each other using their pseudo-labels, and we ensemble their predictions to improve overall performance. Extensive experiments on various benchmarks show that our method consistently outperforms state-of-the-art baselines.Code and appendix can be found in https://github.com/YukiNozzzz/DEACON.

Bayesian optimization is widely employed for optimizing complex black-box functions but struggles with the curse of dimensionality. Random embedding, as a dimension reduction strategy, simplifies tasks that possess the effective dimension by optimizing within a low-dimensional subspace. However, determining the effective dimension of a task in advance remains a significant challenge, which influences the selection of the subspace dimensionality and the optimization performance. Traditional methods use fixed subspace dimensions provided by experts or rely on trial and error to estimate subspace dimensions with resources consumed. To this end, this paper proposes an automated random embedding for high-dimensional Bayesian optimization with unknown effective dimension, called Dynamic Shared Embedding Bayesian Optimization (DSEBO). DSEBO starts with a low dimension and switches to a higher subspace if the solutions in the current subspace show preliminary convergence. DSEBO dynamically determines the dimension of the next subspace based on the quality of the solutions in different subspaces and shares the queried solutions with the new subspace for a better initialization. Theoretically, we derive a regret bound for DSEBO and demonstrate that DSEBO can better balance approximation and optimization errors. Extensive experiments on functions with dimensionality of varying magnitudes and real-world tasks with unknown effective dimensions reveal that, compared with state-of-the-art methods, alternating optimization across different subspaces results in significant improvements in high-dimensional optimization, both in terms of optimization regret and time.

Data-free knowledge distillation enables model distillation without original training data, which is critical for privacy-sensitive tabular domains. However, existing methods do not perform well on tabular data because they do not explicitly address feature interactions, which are critical for encoding predictive knowledge. We identify interaction diversity, systematic coverage of feature combinations, as an important factor for effective tabular model distillation. To operationalize this insight, we propose TabKD, which learns adaptive feature bins aligned with teacher decision boundaries, then generates synthetic queries that ensure uniform pairwise interaction coverage. Across 4 benchmark datasets and 4 teacher architectures, TabKD achieves the highest student-teacher agreement in 14 of 16 configurations, outperforming 5 state-of-the-art baselines. We further show that interaction coverage strongly correlates with distillation quality, validating our core hypothesis. Our work establishes interaction-focused exploration as a principled framework for tabular model distillation.

Generating high-quality synthetic tabular data, regarding fidelity, diversity, and utility, is crucial for many practical purposes. Recent tabular data synthesis methods, including two-stage generative modeling approaches, have achieved this with nearly perfect fidelity. However, we observe that there remains room for improving diversity, and we show that this limitation arises from an additional constraint imposed on the latent support. Our main contribution is that we effectively eliminate this redundant constraint by directly deriving the objective function from the KL-divergence between the ground-truth density and the generative model used for synthetic sample generation. We empirically demonstrate that our model, which relies on a single prior distribution, significantly improves the quality of the synthetic data, especially in terms of diversity. Our implementation code is available at https://anonymous.4open.science/r/SPT-1EE2/.

Image-text alignment relies on large-scale, well-aligned image-text pairs, yet web-crawled datasets inevitably introduce noisy correspondence. Existing methods typically re-weight training pairs by estimating semantic matching degrees from the model's own similarity predictions, which are inherently unreliable under noise and often overestimate partially aligned pairs. To overcome this limitation, we introduce a novel semantic matching degree estimation method based on ranking consistency. Our basic idea is that for a well-aligned image-text pair (I, T), the image I and text T should occupy semantically consistent positions in the embedding space. Thus, using I and T as queries to retrieve other texts (images) should yield two highly consistent rankings. Therefore, we reweight the training paris by measuring the Normalized Discounted Cumulative Gain (NDCG) between its cross-modal and intra-modal retrieval results. This signal is derived from the global data manifold structure rather than point-wise predictions, making it more robust to noise. Experiments on multiple cross-modal retrieval benchmarks validate the effectiveness of our method. Code is available at https://github.com/Aliinton/RCR.

Self-attention is central to the success of Transformer architectures; however, learning the query, key, and value projections from random initialization remains challenging and computationally expensive. In this paper, we propose two complementary methods that leverage the Discrete Cosine Transform (DCT) to enhance the efficiency and performance of Vision Transformers. First, we address the initialization problem by introducing a simple yet effective DCT-based initialization strategy for self-attention, where projection weights are initialized using DCT coefficients. This structure-preserving approach consistently improves classification accuracy on the CIFAR-10 and ImageNet-1K benchmarks. Second, we propose a DCT-based attention compression technique that exploits the decorrelation properties of the frequency domain. By observing that high-frequency DCT coefficients typically correspond to noise, we truncate high-frequency components of the input patches, thereby reducing the dimensionality of the query, key, and value projections without sacrificing accuracy. Experiments on Swin Transformer models demonstrate that the proposed compression method achieves a substantial reduction in computational overhead while maintaining comparable performance. Code: https://github.com/NUBagciLab/DCT-Transformer.