ICML 2026 Papers

Reinforcement learning (RL) has equipped LLM agents with a strong ability to solve complex tasks. However, existing RL methods normally use a single policy network, causing simplicity bias where simple tasks occupy most parameters and dominate gradient updates, leaving insufficient capacity for complex tasks. A plausible remedy could be employing the Mixture-of-Experts (MoE) architecture in the policy network, as MoE allows different parameters (experts) to specialize in different tasks, preventing simple tasks from dominating all parameters. However, a key limitation of traditional MoE is its token-level routing, where the router assigns each token to specialized experts, which fragments phase-consistent patterns into scattered expert assignments and thus undermines expert specialization. In this paper, we propose Phase-Aware Mixture of Experts (PA-MoE). It first features a lightweight phase router that learns latent phase boundaries directly from the RL objective without pre-defining phase categories. Then, the phase router allocates temporally consistent assignments to the same expert, allowing experts to preserve phase-specific expertise. Experimental results demonstrate the effectiveness of our proposed PA-MoE. Code is available at https://anonymous.4open.science/r/PA-MoE-576C/.

Retrieval-Augmented Generation (RAG) systems deployed in sensitive domains must provide interpretable evidence selection and robust safeguards against data poisoning, yet current approaches rely on opaque similarity-based retrieval with arbitrary top-k cutoffs that offer no explanation for their selections and remain vulnerable to adversarial manipulation. We propose METEORA, a rationale-driven RAG framework that addresses these fundamental limitations through interpretable, adaptive evidence retrieval. Our framework introduces three synergistic contributions. First, we preference-tune a general-purpose LLM to generate explicit rationales that articulate why specific evidence is needed for a given query. These rationales then guide adaptive evidence selection through a two-step process: rationale-chunk pairing for query-specific relevance assessment, followed by dynamic cutoff detection that eliminates the need for arbitrary k heuristics. Finally, the same rationales enable a verification stage that filters poisoned or misleading evidence before generation. Evaluation across six datasets demonstrates substantial improvements on three critical dimensions. For retrieval quality, METEORA achieves **21.05\%** higher precision than the best-performing baseline, while its variant with context expansion achieves **13.41\%** higher recall. In terms of efficiency, the framework reduces the volume of evidence required to reach comparable recall by **80\%**, which directly translates to a **33.34\%** improvement in downstream answer generation accuracy. Most notably for adversarial robustness, METEORA increases the F1 score from **0.10 to 0.44** under poisoning attacks, a 4.4$\times$ improvement that makes RAG systems substantially more resilient to adversarial manipulation. Human evaluation with four experienced annotators confirms genuine interpretability, achieving a mean confidence score of **3.64/5** and demonstrating that humans can reliably reconstruct evidence-level decisions with **86\% accuracy**. These results demonstrate that rationale-driven retrieval can simultaneously enhance interpretability, efficiency, and safety in RAG systems for sensitive domains. The code is available in the anonymous GitHub repository \url{https://anonymous.4open.science/r/METEORA-DC46/README.md}

Automatic loop invariant generation remains a challenging problem in program verification, particularly for memory-manipulating programs where shape invariants are required to characterize heap-allocated structures and memory layouts. While existing approaches succeed on numerical invariants, they achieve limited accuracy on shape invariants. We hypothesize that this stems from the need to reason about memory state evolution—information that remains implicit in source code. To address this, we ground LLM reasoning in symbolic execution traces that explicitly capture such transitions. We propose \textsc{Malice}, a two-stage framework incorporating these traces: (1) guided multi-step reasoning that predicts invariants via chain-of-thought analysis of traces, and (2) agentic iterative refinement that corrects candidates through verification tool feedback. Evaluated on LIG-MM+, a benchmark featuring common operations on typical memory structures, \textsc{Malice} substantially outperforms existing approaches.

Real-world user requests to LLM agents are often underspecified. Agents must interact to acquire missing information and make correct downstream decisions. However, current multi-turn GRPO-based methods often rely on trajectory-level reward computation, which leads to credit assignment problems and insufficient advantage signals within rollout groups. A feasible approach is to identify valuable interaction turns at a fine granularity to drive more targeted learning. To address this, we introduce InfoPO, which frames multi-turn interaction as a process of active uncertainty reduction and computes an information-gain reward that credits turns whose feedback measurably changes the agent’s subsequent action distribution compared to a masked-feedback counterfactual. It then combines this signal with task outcomes via an adaptive variance-gated fusion to identify information importance while maintaining task oriented goal direction. Across diverse tasks including intent clarification, collaborative coding, and tool-augmented decision making, InfoPO consistently outperforms prompting and multi-turn RL baselines. It also demonstrates robustness under user simulator shifts and generalizes effectively to environment interactive tasks. Overall, InfoPO provides a principled and scalable mechanism for optimizing complex agent user collaboration.

Recent unified models integrate multimodal understanding and generation within a single framework. However, an ``understanding-generation gap'' persists, where models can capture user intent but often fail to translate this semantic knowledge into precise pixel-level manipulation. This gap results in two bottlenecks in anything-to-image task~(X2I): the \textbf{attention entanglement bottleneck}, where blind planning struggles with complex prompts, and the \textbf{visual refinement bottleneck}, where unstructured feedback fails to correct imperfections efficiently. In this paper, we propose a novel framework that empowers unified models to autonomously switch between generation strategies based on instruction complexity and model capability. To achieve this, we construct a hierarchical data pipeline that constructs execution paths across three adaptive modes: direct generation for simple cases, self-reflection for quality refinement, and multi-step planning for decomposing complex scenarios. Building on this pipeline, we contribute a high-quality dataset with over 50,000 samples and implement a two-stage training strategy comprising SFT and RL. Specifically, we design step-wise reasoning rewards to ensure logical consistency and intra-group complexity penalty to prevent redundant computational overhead. Extensive experiments demonstrate that our method outperforms existing baselines on X2I, achieving superior generation fidelity among simple-to-complex instructions.

Linear attention reformulates sequence modeling as recurrent state evolution, enabling efficient linear-time inference. Under the key–value associative paradigm, existing approaches restrict the role of the query to the readout operation, decoupling it from state evolution. We show that query-conditioned state readout induces a structured value prediction over accumulated memory that complements key-based retrieval. Based on this insight, we propose Q-Delta, a query-aware delta rule that integrates mixed key--query prediction errors into state evolution, enabling jointly corrective dynamics while preserving delta-rule efficiency. We establish stability guarantees for the resulting dynamics and derive a hardware-efficient chunkwise-parallel formulation with a custom Triton implementation. Empirical results demonstrate stable optimization, competitive throughput, and consistent improvements over strong baselines on language modeling and long-context retrieval tasks. Code is available at \url{https://github.com/psmiz/Q-Delta}.

We investigate why deep neural networks suffer from loss of plasticity in continual learning, and thus fail to learn new tasks without reinitializing parameters. We show that this failure is preceded by Hessian spectral collapse at new-task initialization, where meaningful curvature directions vanish and gradient descent becomes ineffective. Analyzing a linearized ReLU network, we derive explicit $\epsilon$-rank conditions for successful training and prove that the loss-weighted Gram matrix is spectrally equivalent to the Generalized Gauss-Newton approximation, thereby relating NTK dynamics to Hessian curvature. Targeting spectral collapse directly, we then discuss the Kronecker factored approximation of the Hessian, which motivates two regularization enhancements: maintaining high effective feature rank and applying L2 penalties. Experiments on continual supervised and reinforcement learning tasks confirm that combining these two regularizers effectively preserves plasticity.

The maximal coding rate reduction ($\text{MCR}^2$) objective is proposed for learning low-dimensional subspace representations and for principled deep model design, where layer structures are derived by unrolling its optimization steps. However, existing methods motivated by this objective do not fully adhere to design principles implied by the $\text{MCR}^2$ gradient, which weakens the principled and interpretable foundations of the resulting models. In this work, we introduce PACEAttention, a novel principled attention mechanism inspired by the *geometric insight* of $\text{MCR}^2$, whose gradient-based updates move features along directions shaped by the underlying low-dimensional feature structure. Our method captures this structure by leveraging randomization to guide feature updates. This principled construction enables the resulting PACENet to exhibit enhanced interpretability, with different heads attending to distinct image regions and capturing *fine-grained* structures under simple supervised training. Experiments demonstrate that PACEAttention achieves superior performance and more stable scalability than previous principled modules while remaining low complexity.

To define a steepest descent method over a neural network, we need to choose a norm for each layer, a way to aggregate these norms across layers, and whether to use normalization. We systematically explore different alternatives for aggregating norms across layers, both formalizing existing combinations of Adam and the recently proposed Muon as a type of non-Euclidean gradient descent, and deriving new variants of the Muon optimizer. Through a comprehensive experimental evaluation of the optimizers within our framework, we find that Muon is sensitive to the choice of learning rate, whereas a new variant we call MuonMax is significantly more robust. We then show how to combine any non-Euclidean gradient method with model based momentum (known as Momo). The new Momo variants of Muon are significantly more robust to hyperparameter tuning, and often achieve a better validation score. Thus for new tasks, where the optimal hyperparameters are not known, we advocate for using Momo in combination with MuonMax to save on costly hyperparameter tuning.

Multimodal time series forecasting has garnered significant attention for its potential to provide more robust and accurate predictions than traditional single-modality models by leveraging rich information inherent in other modalities. However, due to fundamental challenges in modality alignment, existing methods often struggle to effectively incorporate multimodal data into predictions, particularly textual information that has a causal influence on time series fluctuations, such as emergency reports and policy announcements. In this paper, we reflect on the role of textual information in numerical forecasting and propose **Ti**me series transformers with Multimodal **Mi**xture-of-Experts, **TiMi**, to unleash the causal reasoning capabilities of LLMs. Concretely, TiMi utilizes language models to generate inferences on future developments, which then serve as guidance for time series forecasting. To seamlessly integrate both exogenous factors and time series into predictions, we introduce a Multimodal Mixture-of-Experts (MMoE) module as a lightweight plug-in to empower Transformer-based time series models for multimodal forecasting, eliminating the need for explicit representation-level alignment. Experimentally, our proposed TiMi demonstrates consistent state-of-the-art performance on sixteen real-world multimodal forecasting benchmarks, outperforming advanced baselines while offering strong adaptability and interpretability.

We study the problem of monitoring model performance in dynamic environments where labeled data are limited. To this end, we propose prediction-powered risk monitoring (PPRM), a semi-supervised risk-monitoring approach based on prediction-powered inference (PPI). PPRM constructs anytime-valid lower bounds on the running risk by combining synthetic labels with a small set of true labels. Harmful shifts are detected via a threshold-based comparison with an upper bound on the nominal risk, satisfying assumption-free finite-sample guarantees on the type-I error. We demonstrate the effectiveness of PPRM through extensive experiments on image classification, large language model (LLM), and telecommunications monitoring tasks.

Chain-of-thought (CoT) reasoning has proven effective for enhancing problem-solving in large language models. However, when applied to multimodal LLMs (MLLMs), existing CoT approaches suffer from a fundamental limitation: they perform reasoning entirely in text without accessing visual features during the reasoning process. After initial visual encoding, image information becomes inaccessible, forcing models to reason based solely on whatever was captured in the initial description, which forms a `vision-blind reasoning' paradigm that limits fine-grained visual extraction, error verification, and adaptive attention. We propose Text-Visual Interleaved Chain-of-Thought (TVI-CoT), a framework that enables explicit interleaving of textual reasoning and visual feature access through learnable control tokens ($\langle\text{Think}\rangle$, $\langle\text{Look}\rangle$, $\langle\text{Answer}\rangle$). These tokens allow dynamic switching between reasoning and visual grounding, attending to relevant image regions conditioned on the evolving reasoning state. Experiments on eight benchmarks demonstrate state-of-the-art results among MLLM-based CoT methods and notable performance boost compared to the baseline: +6.1\% on MMMU, +3.8\% on MathVerse, +3.4\% on MathVista, and +3.4\% on ScienceQA. Code is available at https://github.com/hulianyuyy/TVI-CoT.

*De novo* functional protein design aims to generate protein sequences that realize specified biochemical functions without relying on evolutionary templates, enabling broad applications in biotechnology and medicine. Existing approaches adopt either direct function-to-sequence mapping or decoupled structure-sequence generation strategies but often fail to achieve functionality and foldability simultaneously. To address this, we propose **CodeFP**, a **Co**-generative protein language model for ***de** novo* **F**unctional **P**rotein design that simultaneously decodes sequence and structure tokens, thereby enabling superior simultaneous realization of functionality and foldability. CodeFP utilizes functional local structures to enrich functional semantic encodings, overcoming the suboptimal translation of flat encodings into structure tokens, while introducing auxiliary functional supervision to alleviate training ambiguity stemming from the one-to-many structure-to-token mapping. Extensive experiments show that CodeFP consistently achieves average improvements of 6.1\% in functional consistency and 3.2\% in foldability over the strongest baseline.

Time series is a pervasive data type across various application domains, rendering the reasonable solving of diverse time series tasks a long-standing goal. Recent advances in large language models (LLMs), especially their reasoning abilities unlocked through reinforcement learning (RL), have opened new opportunities for tackling tasks with long Chain-of-Thought (CoT) reasoning. However, leveraging LLM reasoning for time series remains infancy, hindered by the absence of carefully curated time series CoT data for training, limited data efficiency caused by underexplored data scheduling, and the lack of RL algorithms tailored for exploiting such time series CoT data. In this paper, we introduce VeriTime, a framework that tailors LLMs for time series reasoning through data synthesis, data scheduling, and RL training. First, we propose a data synthesis pipeline that constructs a time series–text multimodal dataset with process-verifiable annotations. Second, we design a data scheduling mechanism that arranges training samples according to a principled hierarchy of difficulty and task taxonomy. Third, we develop a two-stage reinforcement finetuning featuring fine-grained, multi-objective rewards that leverages verifiable process-level CoT data. Extensive experiments show that VeriTime substantially boosts LLM performance across diverse time series reasoning tasks. Notably, it enables compact 3B–4B models to achieve reasoning capabilities on par with or exceeding those of larger proprietary LLMs.

Existing language-image pre-training for remote sensing object detection is constrained by Monolithic Label Learning, which relies on exhaustively enumerating open-set categories via black-box data to acquire fine-grained representations, creating a dependency incompatible with the domain's inherent data scarcity. To transcend this bottleneck, we propose SLIP-RS, establishing a Structured-Attribute Decoupling Paradigm that maps the open-ended category space into a finite, physically meaningful attribute space, unlocking fine-grained discriminability via explicit structural logic. This paradigm is realized via two technical pillars: (1) Structured-Attribute Contrastive Learning, which enforces the learning of decoupled intrinsic visual logic via combinatorial attribute augmentation; and (2) Conformal Attribute Reliability Engine, which leverages conformal prediction theory to rigorously distill high-fidelity supervision from noisy sources, yielding RS-Attribute-15M, the largest dataset with over 15 million attribute annotations. Extensive experiments demonstrate that SLIP-RS establishes unprecedented performance in fine-grained detection and cross-domain generalization, validating structured attributes as a vital foundation for scalable remote sensing models.

Generative modeling of neural network parameters is often tied to architectures because standard parameter representations rely on known weight-matrix dimensions. Generation is further complicated by permutation symmetries that allow networks to model similar input-output functions while having widely different, unaligned parameterizations. In this work, we introduce Neural Network Diffusion Transformers (NNiTs), which generate weights in a width-agnostic manner by tokenizing weight matrices into patches and modeling them as locally structured fields. We establish that Graph HyperNetworks (GHNs) with a convolutional neural network (CNN) decoder structurally align the weight space, creating the local correlation necessary for patch-based processing. Focusing on Multilayer Perceptrons (MLPs), where permutation symmetry is especially apparent, NNiTs generate fully functional networks across a range of architectures. Our approach jointly models discrete architecture tokens and continuous weight patches within a single sequence model. On ManiSkill3 robotics tasks, NNiT achieves $>85\%$ success on architecture topologies unseen during training, while baseline approaches fail to generalize; the same pipeline also generalizes to MNIST classification beyond the robotic control setting.

Bayesian optimization (BO) selects evaluation points for expensive black-box objectives using Gaussian process (GP) predictive distributions. Kernel choice and hyperparameter selection can lead to miscalibrated predictive distributions and an inappropriate exploration--exploitation trade-off. For minimization, sampling criteria such as expected improvement (EI) depend on the predictive distribution below the current best value, so lower-tail miscalibration directly affects the sampling decision. This article studies goal-oriented calibration of GP predictive distributions below a low threshold $t$ in the noiseless setting, for standard GP models with hyperparameters selected by maximum likelihood. A framework for predictive reliability below $t$ is introduced, based on two notions of spatial calibration: occurrence calibration over the design space and thresholded $\mu$-calibration on sublevel sets of the form $\lbrace x\in\mathbb{X}, f(x)\le t \rbrace$. Building on this framework, we propose tcGP, a post-hoc method that calibrates GP predictive distributions below $t$, and we show that the resulting EI-based global optimization algorithm remains dense in the design space. Experiments on standard benchmarks show improved lower-tail calibration and BO performance relative to standard GP models and globally calibrated GP models.

Code Large Language Models (CLLMs) serve as the core of modern code agents, enabling developers to automate complex software development tasks. In this paper, we present Poison-with-Style (PwS), a practical and stealthy model poisoning attack targeting CLLMs. Unlike prior attacks that assume an active adversary capable of directly embedding explicit triggers (e.g., specific words) into developers' prompts during inference, PwS leverages developers' code styles as covert triggers implicitly embedded within their prompts. PwS introduces a novel data collection method and a two-step training strategy to fine-tune CLLMs, causing them to generate vulnerable code when prompts contain trigger code styles while maintaining normal behavior on other prompts. Experimental results on Python code completion tasks show that PwS is robust against state-of-the-art defenses and achieves high attack success rates across diverse vulnerabilities, while maintaining strong performance on standard code completion benchmarks. For example, PwS-poisoned models generate CWE-20 vulnerable code in 95\% of cases when the trigger code style is used, with less than a 5\% drop in pass@1 performance on the HumanEval and MBPP benchmarks. Our implementation and dataset are here: https://github.com/khangtran2020/pws.

Vision language models (VLMs) often generate hallucination, i.e., content that cannot be substantiated by either textual or visual inputs. Prior work primarily attributes this to over-reliance on linguistic prior knowledge rather than visual inputs. Some methods attempt to mitigate hallucination by amplifying visual token attention proportionally to their attention scores. However, these methods overlook the visual attention sink problem, where attention is frequently misallocated to task-irrelevant visual regions, and neglect cross-modal fusion balance by enhancing only visual attention without adjusting attention to the user query. This can result in amplifying incorrect areas while failing to properly interpret the user query. To address these challenges, we propose a simple yet effective method called Gaze Shift-Guided Cross-modal Fusion Enhancement (GIFT). GIFT pre-computes a holistic visual saliency map by tracking positive changes in visual attention, or "gaze shifts", during user query comprehension, and leverages this map to amplify attention to both salient visual information and the user query at each decoding step. This reduces the impact of visual attention sink, as irrelevant tokens exhibit minimal shifts, while ensuring balanced cross-modal fusion for well-integrated representation. Extensive experiments show that GIFT effectively mitigates hallucination in VLMs across both generative and classification tasks, achieving up to 20.7% improvement over greedy decoding, while maintaining general vision-language performance with low computational overhead.