ICLR 2026 Papers

A central objective of manipulation policy design is to enable robots to comprehend human instructions and predict generalized actions in unstructured environments. Recent autoregressive vision-language-action (VLA) approaches discretize actions into bins to exploit the pretrained reasoning and generation paradigms of vision-language models (VLMs). While these models achieve efficient and scalable training, the discretization undermines the continuity required for precise control. In contrast, diffusion-based VLA methods incorporate an additional diffusion head to predict continuous actions, but they rely solely on feature representations extracted from the VLM, without leveraging the pretrained large language model (LLM) as an expert for iterative action generation. To integrate the complementary strengths of autoregressive and diffusion generation, we introduce HybridVLA, which innovatively leverages a shared LLM backbone to perform iterative action prediction through both paradigms. Specifically, a collaborative training recipe is proposed, incorporating diffusion denoising into the next-token prediction process and mitigating interference between the two generation paradigms. With this recipe, we find these two action prediction methods not only reinforce each other but also exhibit varying strengths across different scenarios. Therefore, we design a collaborative action ensemble mechanism that adaptively fuses both predictions, leading to more robust control. HybridVLA outperforms previous state-of-the-art VLA methods by 17\% and 19\% in mean success rate on simulation and real-world tasks, respectively, while demonstrating generalization to unseen configurations.

Subject-agnostic brain decoding, which aims to reconstruct continuous visual experiences from fMRI without subject-specific training, holds great potential for clinical applications. However, this direction remains underexplored due to challenges in cross-subject generalization and the complex nature of brain signals. In this work, we propose Visual Cortex Flow Architecture (VCFlow), a novel hierarchical decoding framework that explicitly models the ventral-dorsal architecture of the human visual system to learn multi-dimensional representations. By disentangling and leveraging features from early visual cortex, ventral, and dorsal streams, VCFlow captures diverse and complementary cognitive information essential for visual reconstruction. Furthermore, we introduce a feature-level contrastive learning strategy to enhance the extraction of subject-invariant semantic representations, thereby enhancing subject-agnostic applicability to previously unseen subjects. Unlike conventional pipelines that need more than 12 hours of per-subject data and heavy computation, VCFlow sacrifices only 7\% accuracy on average yet generates each reconstructed video in 10 seconds without any retraining, offering a fast and clinically scalable solution.

Autoregressive (AR) models are promising for image generation, yet continuous-token AR variants often trail latent diffusion and masked-generation models. The core issue is heterogeneous variance in VAE latents, which is amplified during AR decoding, especially under classifier-free guidance (CFG), and can cause variance collapse. We propose SphereAR to address this issue. Its core design is to constrain all AR inputs and outputs---including after CFG---to lie on a fixed-radius hypersphere (constant $\ell_2$ norm), leveraging hyperspherical VAEs. Our theoretical analysis shows that hyperspherical constraint removes the scale component (the primary cause of variance collapse), thereby stabilizing AR decoding. Empirically, on ImageNet generation, SphereAR-H (943M) sets a new state of the art for AR models, achieving FID 1.34. Even at smaller scales, SphereAR-L (479M) reaches FID 1.54 and SphereAR-B (208M) reaches 1.92, matching or surpassing much larger baselines such as MAR-H (943M, 1.55) and VAR-d30 (2B, 1.92). To our knowledge, this is the first time a pure next-token AR image generator with raster order surpasses diffusion and masked-generation models at comparable parameter scales.

Advances in training and architectural design have enabled LLMs with million-token context windows, yet in practice these models often read far more than they can reliably use. While inference-time compute scaling—typically via “thinking tokens”—can help on short multi-step reasoning tasks, our controlled long-context experiments show rapidly diminishing returns that collapse as context grows. We trace this to score dilution in static self-attention and prove that, in such regimes, decoding more tokens cannot reliably recover buried evidence. We propose query-only test-time training (qTTT): a cache-preserving adaptation that performs a single prefill to fix keys/values and then applies a handful of gradient updates to the query projections. qTTT provably increases the target–distractor margin and, empirically, delivers consistent gains across model sizes and benchmarks. On Qwen3-4B, qTTT improves average accuracy by +12.6 and +14.1 absolute points on LongBench-v2 and ZeroSCROLLS, respectively. The practical takeaway is simple: for long contexts, spending a small inference-time budget on context-specific adaptation is a more effective use of compute than generating additional thinking tokens.

LLMs can perform seemingly planning-intensive tasks, like writing coherent stories or functioning code, without explicitly verbalizing a plan; however, the extent to which they implicitly plan is unknown. In this paper, we define *latent planning* as occurring when LLMs possess internal planning representations that (1) cause the generation of a specific future token or concept, and (2) shape preceding context to license said future token or concept. We study the Qwen-3 family (0.6B-14B) on simple planning tasks, finding that latent planning ability increases with scale. Models that plan possess features that represent a planned-for word like *accountant*, and cause them to output *an* rather than *a*; moreover, even the less-successful Qwen-3 4B-8B have nascent planning mechanisms. On the more complex task of completing rhyming couplets, we find that models often identify a rhyme ahead of time, but even large models seldom plan far ahead. However, we can elicit some planning that increases with scale when steering models towards planned words in prose. In sum, we offer a framework for measuring planning and mechanistic evidence of how models' planning abilities grow with scale.

In the world of Harry Potter, when Dumbledore's mind is overburdened, he extracts memories into a Pensieve to be revisited later. In the world of AI, while we possess the Pensieve—mature databases and retrieval systems, our models inexplicably lack the "wand" to operate it. They remain like a Dumbledore without agency, passively accepting a manually engineered context as their entire memory. This work finally places the wand in the model's hand. We introduce StateLM, a new class of foundation models endowed with an internal reasoning loop to manipulate their own state. We equip our model with a suite of tools, such as dynamic indexing, context pruning, and note-taking, and train it to actively manage this loop. By learning to dynamically construct its own context, our model breaks free from the architectural prison of a fixed window. The results are prominent: our state-management approach decouples performance from context window size, delivering strong accuracy and sustainability under extremely long contexts with linear inference cost. We demonstrate this by showing StateLM reliably retrieves a "needle" from a 1-million-token haystack, a task far beyond the reach of conventional models. On practical document QA tasks from NovelQA and $\infty$Bench, StateLM outperforms strong instruct baselines while using only 1/4 of their active context. An ablation further shows that our curated training pipeline is more effective for learning memory management than agent-like prompting. Together, these results mark a shift from passive predictors to state-aware systems where reasoning becomes a stateful and manageable process.

There are two primary approaches to solving Markov decision problems (MDPs): dynamic programming based on the Bellman equation and linear programming (LP). Dynamic programming methods are the most widely used and form the foundation of both classical and modern reinforcement learning (RL). By contrast, LP-based methods have been less commonly employed, although they have recently gained attention in contexts such as offline RL. The relative underuse of the LP-based methods stems from the fact that it leads to an inequality-constrained optimization problem, which is generally more challenging to solve effectively compared with Bellman-equation-based methods. The purpose of this paper is to establish a theoretical foundation for solving LP-based MDPs in a more effective and practical manner. Our key idea is to leverage the log-barrier function, widely used in inequality-constrained optimization, to transform the LP formulation of the MDP into an unconstrained optimization problem. This reformulation enables approximate solutions to be obtained easily via gradient descent. While the method may appear naive, to the best of our knowledge, a thorough theoretical interpretation of this approach has not yet been developed. This paper aims to bridge this gap.

We introduce Cautious Weight Decay (CWD), a one-line, optimizer-agnostic modification that applies weight decay only to parameter coordinates whose signs align with the optimizer update. Unlike standard decoupled decay, which implicitly optimizes a regularized or constrained objective, CWD preserves the original loss and admits a bilevel interpretation: it induces sliding-mode behavior upon reaching the stationary manifold, allowing it to search for locally Pareto-optimal stationary points of the unmodified objective. In practice, CWD is a drop-in change for optimizers such as AdamW, Lion, and Muon, requiring no new hyperparameters or additional tuning. For language model pre-training and ImageNet classification, CWD consistently improves final loss and accuracy at million- to billion-parameter scales.

Vision-Language Models (VLMs) demonstrate impressive performance in understanding visual content with language instruction by converting visual inputs to vision tokens. However, redundancy in vision tokens results in the degenerated inference efficiency of VLMs. While many algorithms have been proposed to reduce the number of vision tokens, most of them apply only unimodal information (i.e., vision/text) for pruning and ignore the inherent multimodal property of vision-language tasks. Moreover, it lacks a generic criterion that can be applied to different modalities. To mitigate this limitation, in this work, we propose to leverage both vision and text tokens to select informative vision tokens by the coverage criterion. We first formulate the subset selection problem as a maximum coverage problem. Afterwards, a subset of vision tokens is optimized to cover the text tokens and the original set of vision tokens, simultaneously. The proposed method MMTok is extensively evaluated on benchmark datasets with different VLMs. The comparison illustrates that vision and text information are complementary, and combining multimodal information can surpass the unimodal baseline with a clear margin. Moreover, under the maximum coverage criterion on the POPE dataset, our method achieves a 1.87× speedup while maintaining 98.7\% of the original performance on LLaVA-NeXT-13B. Furthermore, with only four vision tokens, 87.7\% of the original performance is still preserved on LLaVA-1.5-7B. These results highlight the effectiveness of coverage in token selection.

Multiplex Biological Networks (MBNs), which represent multiple interaction types between entities, are crucial for understanding complex biological systems. Yet, existing methods often inadequately model multiplexity, struggle to integrate structural and sequence information, and face difficulties in zero-shot prediction for unseen entities with no prior neighbourhood information. To address these limitations, we propose a novel framework for zero-shot interaction prediction in MBNs by leveraging context-aware representation learning and knowledge distillation. Our approach leverages domain-specific foundation models to generate enriched embeddings, introduces a topology-aware graph tokenizer to capture multiplexity and higher-order connectivity, and employs contrastive learning to align embeddings across modalities. A teacher–student distillation strategy further enables robust zero-shot generalization. Experimental results demonstrate that our framework outperforms state-of-the-art methods in interaction prediction for MBNs, providing a powerful tool for exploring various biological interactions and advancing personalized therapeutics.

Numerous studies on low-rank adaptation (LoRA) emerged in recent years, with the aim of accelerating the convergence of the LoRA framework. In this paper, we leverage the horizontal lift theory from differential geometry to establish the general iteration scheme on the quotient manifold \mathbb{R}\_\*^{m \times r} \times \mathbb{R}\_\*^{n \times r}/\sim. By endowing the LoRA framework with Riemannian quotient geometries, our theory not only guarantees efficient feature learning but also bridges the LoRA algorithms and the pre-training algorithms for large models. Furthermore, we theoretically analyze the role of the weight decay matrix $\epsilon_{decay}I$ in efficient feature learning and then replace it with the Sylvester matrix $K$, indicating that the theory helps remove an important hyperparameter while generating accurate and computationally efficient optimizers. Based on the general scheme, we propose two efficient LoRA optimizers with runtime analysis, Adam-Sylvester (AdamS) and LRACS, then conduct experiments on the transformer-based networks. The results demonstrate evident improvements over existing optimizers.

Unified Vision–Language Models (UVLMs) aim to advance multimodal learning by supporting both understanding and generation within a single framework. However, existing approaches largely focus on architectural unification while overlooking the need for explicit interaction between the two capabilities during task solving. As a result, current models treat understanding and generation as parallel skills rather than synergistic processes. To achieve real synergy, we introduce the interleaved Analyzing–Drafting problem-solving loop (AD-Loop), a new think paradigm that dynamically alternates between analytic and drafting operations. By interleaving textual thoughts with visual thoughts, AD-Loop enables models to iteratively refine both comprehension and outputs, fostering genuine synergy. To train this mechanism, we design a two-stage strategy: supervised learning on interleaved thought data to initialize alternation, followed by reinforcement learning to promote adaptive and autonomous control. Extensive experiments demonstrate that AD-Loop consistently improves performance across standard benchmarks for both understanding and generation, with strong transferability to various UVLMs architectures. Visual analyses further validate the effectiveness of implicit visual thoughts. These results highlight AD-Loop as a principled and broadly applicable strategy for synergizing comprehension and creation.

Robot manipulation policies, while central to the promise of physical AI, are highly vulnerable in the presence of external variations in the real world. Diagnosing these vulnerabilities is hindered by two key challenges: (i) the relevant variations to test against are often unknown, and (ii) direct testing in the real world is costly and unsafe. We introduce a framework that tackles both issues by learning a separate deep reinforcement learning (deep RL) policy for vulnerability prediction through virtual runs on a continuous vision-language embedding trained with limited success-failure data. By treating this embedding space, which is rich in semantic and visual variations, as a potential field, the policy learns to move toward vulnerable regions while being repelled from success regions. This vulnerability prediction policy, trained on virtual rollouts, enables scalable and safe vulnerability analysis without expensive physical trials. By querying this policy, our framework builds a probabilistic vulnerability-likelihood map. Experiments across simulation benchmarks and a physical robot arm show that our framework uncovers up to 23\% more unique vulnerabilities than state-of-the-art vision-language baselines, revealing subtle vulnerabilities overlooked by heuristic testing. Additionally, we show that fine-tuning the manipulation policy with the vulnerabilities discovered by our framework improves manipulation performance with much less fine-tuning data.

Unknown unknowns (U2s) are deployment-time scenarios absent from development/testing. Unlike conventional anomalies, U2s are not out-of-distribution (OOD); they stem from changes in underlying system dynamics without a distribution shift from normal data. Thus, existing multi-variate time series anomaly detection (MTAD) methods—which rely on distribution-shift cues—are ill-suited for U2 detection. Specifically: (i) we show most anomaly datasets exhibit distribution shift between normal and anomalous data and therefore are not representative of U2s; (ii) we introduce eight U2 benchmarks where training data contain OOD anomalies but no U2s, while test sets contain both OOD anomalies and U2s; (iii) we demonstrate that state-of-the-art (SOTA) MTAD results often depend on impractical enhancements: point adjustment (PA) (uses ground truth to flip false negatives to true positives, inflating precision) and threshold learning with data leakage (TL) (tuning thresholds on test data and labels); (iv) with PA+TL, even untrained deterministic methods can match or surpass MTAD baselines; (v) without PA/TL, existing MTAD methods degrade sharply on U2 benchmarks. Finally, we present sparse model identification–enhanced anomaly detection (SPIE-AD), a model-recovery-and-conformance, zero-shot MTAD approach that outperforms baselines on all eight U2 benchmarks and on six additional real-world MTAD datasets—without PA or TL.

Scalable robot policy pre-training has been hindered by the high cost of collecting high-quality demonstrations for each platform. In this study, we address this issue by uniting offline reinforcement learning (offline RL) with cross-embodiment learning. Offline RL leverages both expert and abundant suboptimal data, and cross-embodiment learning aggregates heterogeneous robot trajectories across diverse morphologies to acquire universal control priors. We perform a systematic analysis of this offline RL and cross-embodiment paradigm, providing a principled understanding of its strengths and limitations. To evaluate this offline RL and cross-embodiment paradigm, we construct a suite of locomotion datasets spanning 16 distinct robot platforms. Our experiments confirm that this combined approach excels at pre-training with datasets rich in suboptimal trajectories, outperforming pure behavior cloning. However, as the proportion of suboptimal data and the number of robot types increase, we observe that conflicting gradients across morphologies begin to impede learning. To mitigate this, we introduce an embodiment-based grouping strategy in which robots are clustered by morphological similarity and the model is updated with a group gradient. This simple, static grouping substantially reduces inter-robot conflicts and outperforms existing conflict-resolution methods.

Deep learning models achieve strong performance across various domains but often rely on spurious correlations, making them vulnerable to distribution shifts. This issue is particularly severe in subpopulation shift scenarios, where models struggle in underrepresented groups. While existing methods have made progress in mitigating this issue, their performance gains are still constrained. They lack a theoretical \red{motivation} connecting the embedding space representations with worst-group error. To address this limitation, we propose Spurious Correlation-Aware Embedding Regularization for Worst-Group Robustness (SCER), a novel approach that directly regularizes feature representations to suppress spurious cues. We theoretically show that worst-group error is influenced by how strongly the classifier relies on spurious versus core directions, as identified from differences in group-wise mean embeddings across domains and classes. By imposing theoretical constraints at the embedding level, SCER encourages models to focus on core features while reducing sensitivity to spurious patterns. \blue{Through systematic evaluation on multiple vision and language tasks, we show that SCER outperforms prior state-of-the-art methods in worst-group accuracy.} Our code is available at \href{https://anonymous.4open.science/r/SCER_repo/}{https://anonymous.4open.science/r/SCER}.

We propose a novel class of neural differential equation models called mean-field continuous sequence predictors (MFPs) for efficiently generating continuous sequences with potentially infinite-order complexity. To address complex inductive biases in time-series data, we employ mean-field dynamics structured through carefully designed graphons. By reframing continuous sequence prediction as mean-field games, we utilize a fictitious play strategy integrated with gradient-descent techniques. This approach exploits the stochastic maximum principle to determine the Nash equilibrium of the system. Both empirical evidence and theoretical analysis highlight the unique advantages of our framework, where a collective of continuous predictors achieves highly accurate predictions and consistently outperforms benchmark prior works.

Large language models excel at reasoning but lack key aspects of introspection, including the ability to anticipate their own success and the computation required to achieve it. Humans use real-time introspection to decide how much effort to invest, when to make multiple attempts, when to stop, and when to signal success or failure. Without this ability, LLMs struggle to make intelligent meta-cognition decisions. Test-time scaling methods such as Best-of-N drive up cost and latency by using a fixed budget of samples regardless of the marginal benefit of each one at any point in generation, and the absence of confidence signals can mislead people, prevent appropriate escalation to better tools, and undermine trustworthiness. Learned verifiers or reward models can provide confidence estimates, but do not enable adaptive inference and add substantial inference cost by requiring extra models or forward passes. We present ZIP-RC, an adaptive inference method that equips models with zero-overhead inference-time predictions of reward and cost. At every token during generation, ZIP-RC reuses reserved or unused logits in the same forward pass as next-token prediction to output a joint distribution over final reward and remaining length—no extra models, architecture change, or inference overhead. This full joint distribution is used to compute a sampling utility which is the linear combination of the expected maximum reward, total compute, and latency of set of samples if generated to completion. During inference, we maximize this utility with meta-actions that determine which prefix of tokens to continue or initiate sampling from. On mixed-difficulty mathematical benchmarks, ZIP-RC improves accuracy by up to 12% over majority voting at equal or lower average cost, and traces smooth Pareto frontiers between quality, compute, and latency. By providing real-time reward–cost introspection, ZIP-RC allows models to reason adaptively and more efficiently.

Cross-modal retrieval is a significant task that aims to learn the semantic correspondence between visual and textual modalities. Unsupervised hashing methods can efficiently manage large-scale data and can be effectively applied to cross-modal retrieval studies. However, existing methods typically fail to fully exploit the hierarchical structure between text and image data. Moreover, the commonly used direct modal alignment cannot effectively bridge the semantic gap between these two modalities. To address these issues, we introduce a novel Hierarchical Encoding Tree with Modality Mixup (HINT) method, which achieves effective cross-modal retrieval by extracting hierarchical cross-modal relations. HINT constructs a cross-modal encoding tree guided by hierarchical structural entropy and generates proxy samples of text and image modalities for each instance from the encoding tree. Through the curriculum-based mixup of proxy samples, HINT achieves progressive modal alignment and effective cross-modal retrieval. Furthermore, we conduct cross-modal consistency learning to achieve global-view semantic alignment between text and image representations. Extensive experiments on a range of cross-modal retrieval datasets demonstrate the superiority of HINT over state-of-the-art methods.