ICLR 2026 Papers

Although probing frozen models has become a standard evaluation paradigm, self-supervised learning in audio defaults to fine-tuning when pursuing state-of-the-art on AudioSet. A key reason is that global pooling creates an information bottleneck causing linear probes to misrepresent the embedding quality: The $\texttt{cls}$-token discards crucial token information about dispersed, localized events in multi-label audio. This weakness is rooted in the mismatch between the pretraining objective (operating globally) and the downstream task (localized events). Across a comprehensive benchmark of 13 datasets and 6 spectrogram-based encoders, we first investigate the global pooling bottleneck. We then introduce binarized prototypical probes: a lightweight and simple pooling method that learns prototypes to perform class-wise information aggregation. Despite its simplicity, our method notably outperforms linear and attentive probing. Our work establishes probing as a competitive and efficient paradigm for evaluating audio SSL models, challenging the reliance on costly fine-tuning.

In recent years, integrating multimodal understanding and generation into a single unified model has emerged as a promising paradigm. While this approach achieves strong results in text-to-image (T2I) generation, it still struggles with precise image editing. We attribute this limitation to an imbalanced division of responsibilities. The understanding module primarily functions as a translator that encodes user instructions into semantic conditions, while the generation module must simultaneously act as designer and painter, inferring the original layout, identifying the target editing region, and rendering the new content. This imbalance is counterintuitive because the understanding module is typically trained with several times more data on complex reasoning tasks than the generation module. To address this issue, we introduce *Draw-In-Mind* (DIM), a dataset comprising two complementary subsets: (**i**) DIM-T2I, containing 14M long-context image–text pairs to enhance complex instruction comprehension; and (**ii**) DIM-Edit, consisting of 233K chain-of-thought imaginations generated by GPT-4o, serving as explicit design blueprints for image edits. We connect a frozen Qwen2.5-VL-3B with a trainable SANA1.5-1.6B via a lightweight two-layer MLP, and train it on the proposed DIM dataset, resulting in DIM-4.6B-T2I/Edit. Despite its modest parameter scale, DIM-4.6B-Edit achieves SOTA or competitive performance on the ImgEdit and GEdit-Bench benchmarks, outperforming much larger models such as UniWorld-V1 and Step1X-Edit. These findings demonstrate that explicitly assigning the design responsibility to the understanding module provides significant benefits for image editing. Our dataset and models will be publicly available.

Humans navigate daily life by combining two modes of behavior: deliberate planning in novel situations and fast, automatic responses in familiar ones. Modeling human decision-making therefore requires capturing how people switch between these modes. We present a framework for learning human habits with rule-guided active inference, extending the view of the brain as a prediction machine that minimizes mismatches between expectations and observations, and computationally modeling of human(-like) behavior and habits. In our approach, habits emerge as symbolic rules that serve as compact, interpretable shortcuts for action. To learn these rules alongside the human models, we design a biologically inspired wake--sleep algorithm. In the wake phase, the agent engages in active inference on real trajectories: reconstructing states, updating beliefs, and harvesting candidate rules that reliably reduce free energy. In the sleep phase, the agent performs generative replay with its world model, refining parameters and consolidating or pruning rules by minimizing joint free energy. This alternating rule–model consolidation lets the agent build a reusable habit library while preserving the flexibility to plan. Experiments on basketball player movements, car-following behavior, medical diagnosis, and visual game strategy demonstrate that our framework improves predictive accuracy and efficiency compared to logic-based, deep learning, LLM-based, model-based RL, and prior active inference baselines, while producing interpretable rules that mirror human-like habits.

Graph generation is a critical yet challenging task as empirical analyses require a deep understanding of complex, non-Euclidean structures. Diffusion models have recently made significant achievements in graph generation, but these models are typically adapted from image generation frameworks and overlook inherent higher-order topology, leaving them ill-suited for capturing the topological properties of graphs. In this work, we propose Higher-order Guided Diffusion (HOG-Diff), a principled framework that progressively generates plausible graphs with inherent topological structures. HOG-Diff follows a coarse-to-fine generation curriculum guided by higher-order topology and implemented via diffusion bridges. We further prove that our model exhibits a stronger theoretical guarantee than classical diffusion frameworks. Extensive experiments on both molecular and generic graph generation tasks demonstrate that our method consistently outperforms or remains competitive with state-of-the-art baselines.

Generative models have demonstrated significant results in ultra-low bitrate image compression, owing to their powerful capabilities for content generation and texture completion. Existing works primarily based on diffusion models still face challenges such as limited bitrate adaptability and high computational complexity for encoding and decoding. Inspired by the success of Visual AutoRegressive model (VAR), we introduce AutoRegressive-based Progressive Coding (ARPC) for ultra-low bitrate image compression, a progressive image compression framework based on next-scale prediction visual autoregressive model. Based on multi-scale residual vector quantizer, ARPC efficiently encodes the image into multi-scale discrete token maps and controls the bitrates by selecting different scales for transmission. For decompression, ARPC leverages the prior knowledge inherent in the visual autoregressive model to predict the unreceived scales, which is naturally the autoregressive generation process. To further increase the compression ratio, we target the VAR as a probability estimator for lossless entropy coding and propose group-masked bitwise multi-scale residual quantizer to adaptively allocate bits for different scales. Extensive experiments show that ARPC achieves state-of-the-art perceptual fidelity at ultra-low bitrates and high decompression efficiency compared with existing diffusion-based methods.

Most applications of generative AI involve a sequential interaction in which a person inputs a prompt and waits for a response, and where reaction time and adaptivity are not important factors. In contrast, live jamming is a collaborative interaction that requires real-time coordination and adaptation without access to the other player’s future moves, while preserving diversity to sustain a creative flow. Reinforcement learning post-training enables effective adaptation through on-policy interaction, yet it often reduces output diversity by exploiting coherence-based rewards. This collapse, known as ``reward hacking'', affects many RL post-training pipelines, but is especially harmful in live jamming, where musical creativity relies on dynamic variation and mutual responsiveness. In this paper, we propose a novel adversarial training method on policy-generated trajectories to mitigate reward hacking in RL post-training for melody-to-chord accompaniment. A co-evolving discriminator separates policy trajectories from the data distribution, while the policy maximizes the discriminator output in addition to coherence rewards to prevent collapse to trivial outputs. We evaluate accompaniment quality and output diversity in simulation with both fixed test melodies and learned melody agents, and we conduct a user study with the model deployed in a real-time interactive system with expert musicians. Quantitative evaluation and user feedback demonstrate improved output diversity, harmonic coherence, adaptation speed and user agency. Our results demonstrate a simple yet effective method to mitigate reward hacking in RL post-training of generative sequence models.

Offline reinforcement learning (RL) has achieved notable progress in recent years. However, most existing offline RL methods require a large amount of training data to achieve reasonable performance and offer limited out-of-distribution (OOD) generalization capability due to conservative data-related regularizations. This seriously hinders the usability of offline RL in solving many real-world applications, where the available data are often limited. In this study, we introduce TELS, a highly sample-efficient offline RL algorithm that enables state-stitching in a compact latent space regulated by the fundamental time-reversal symmetry (T-symmetry) of dynamical systems. Specifically, we introduce a T-symmetry enforced inverse dynamics model (TS-IDM) to derive well-regulated latent state representations that greatly facilitate OOD generalization. A guide-policy can then be learned entirely in the latent space to optimize for the reward-maximizing next state, bypassing the conservative action-level behavioral regularization adopted in most offline RL methods. Finally, the optimized action can be extracted using the learned TS-IDM, together with the optimized latent next state from the guide-policy. We conducted comprehensive experiments on both the D4RL benchmark tasks and a real-world industrial control test environment, TELS achieves superior sample efficiency and OOD generalization performance, significantly outperforming existing offline RL methods in a wide range of challenging small-sample tasks.

While foundation models (FMs), such as diffusion models and large vision-language models (LVLMs), have been widely applied in educational contexts, their ability to generate pedagogically effective visual explanations remains limited. Most existing approaches focus primarily on textual reasoning, overlooking the critical role of structured and interpretable visualizations in supporting conceptual understanding. To better assess the visual reasoning capabilities of FMs in educational settings, we introduce EduVisBench, a multi-domain, multi-level benchmark. EduVisBench features diverse STEM problem sets requiring visually grounded solutions, along with a fine-grained evaluation rubric informed by pedagogical theory. Our empirical analysis reveals that existing models frequently struggle with the inherent challenge of decomposing complex reasoning and translating it into visual representations aligned with human cognitive processes. To address these limitations, we propose EduVisAgent, a multi-agent collaborative framework that coordinates specialized agents for instructional planning, reasoning decomposition, metacognitive prompting, and visualization design. Experimental results show that EduVisAgent substantially outperforms all baselines, achieving a 40.2% improvement and delivering more educationally aligned visualizations.

Large language model (LLM) alignment is typically achieved through learning from human preference comparisons, making the quality of preference data critical to its success. Existing studies often pre-process raw training datasets to identify valuable preference pairs using external reward models or off-the-shelf LLMs, achieving improved overall performance but rarely examining whether individual, selected data point is genuinely beneficial. We assess data quality through individual influence on validation data using our newly proposed truncated influence function (TIF), which mitigates the over-scoring present in traditional measures and reveals that preference data quality is inherently a property of the model. In other words, a data pair that benefits one model may harm another. This leaves the need to improve the preference data selection approaches to be adapting to specific models. To this end, we introduce a set of candidate scoring functions (SFs) that are computationally simpler than TIF and positively correlated with it. These functions are also model dependent and can serve as potential indicators of individual data quality for preference data selection. Furthermore, we observe that these SFs inherently exhibit errors when compared to TIF. To this end, we combine them to offset their diverse error sources, resulting in a simple yet effective data selection rule that enables the models to achieve a more precise selection of valuable preference data. We conduct experiments across diverse alignment benchmarks and various LLM families, with results demonstrating that better alignment performance can be achieved using less data, showing the generality and robustness of our findings and new methods.

Large language models (LLMs) can be enhanced with external knowledge through two dominant approaches: (1) **retrieval-augmented generation (RAG)**, which supplements LLMs with in-context retrieved passages, and (2) **parametric knowledge adaptation (PKA)**, which directly updates model parameters with new domain knowledge. Recently, parametric RAG (PRAG) has emerged as a promising framework, extending RAG by translating retrieved passages into parameter updates, thereby mitigating inefficiency and noise sensitivity inherent to RAG. However, existing PRAG methods remain limited to single-pass retrieval, falling short of the \textbf{multi-hop RAG} setting that requires iterative retrieval and reasoning. We propose **MergePRAG**(*Orthogonal Merging of Passage-experts for Multi-hop PRAG*), a novel framework that sequentially integrates retrieved passages into LLM parameters through a continual merging mechanism, which is advanced by two key proposals: (1) **orthogonal merging** using the Gram–Schmidt process to minimize conflicts between "passage experts", and (2) **critical-layer parameterization** to efficiently encode in-context passages. Experiments on multi-hop open-domain QA and reasoning-aware knowledge editing show that MergePRAG consistently outperforms both standard and state-of-the-art RAGs as well as existing parametric adaptation methods, achieving superior effectiveness and efficiency. All datasets and code will be released at {https://anonymous.4open.science/r/MhQA_hypernetwork-B31F.

Driven by the rapid growth of model parameters, parameter-efficient fine-tuning (PEFT) has become essential for adapting large models to diverse downstream tasks under constrained computational resources. Within this paradigm, orthogonal fine-tuning and its variants preserve semantic representations of pre-trained models, but struggle to achieve both expressiveness and efficiency in terms of parameter counts, memory, and computation. To overcome this limitation, we propose efficient Orthogonal Fine-Tuning with Principal Subspace adaptation (PSOFT), which confines orthogonal transformations to the principal subspace of pre-trained weights. Specifically, PSOFT constructs this subspace via matrix decomposition to enable compatible transformations with higher rank, establishes a theoretical condition that strictly maintains the geometry of this subspace for essential semantic preservation, and introduces efficient tunable vectors that gradually relax orthogonality during training to enhance adaptability. Extensive experiments on 35 NLP and CV tasks across four representative models demonstrate that PSOFT offers a practical and scalable solution to simultaneously achieve semantic preservation, expressiveness, and multi-dimensional efficiency in PEFT.

Tool-integrated reasoning has emerged as a key focus for enabling agentic applications. Among these, DeepResearch Agents have gained significant attention for their strong performance on complex, open-ended information-seeking tasks. We introduce Fathom-DeepResearch, an agentic system composed of two specialized models. The first is Fathom-Search-4B, a DeepSearch model trained from Qwen3-4B and optimized for evidence-based investigation through live web search and targeted webpage querying. Its training combines three advances: (i) DUETQA, a ∼5K-sample dataset generated via multi-agent self-play that enforces strict web-search dependence and heterogeneous source grounding; (ii) RAPO, a zero-overhead extension of GRPO that stabilizes multi-turn Reinforcement Learning with Verifiable Rewards through curriculum pruning, reward-aware advantage scaling, and per-prompt replay buffers; and (iii) a steerable step-level reward that classifies each tool call by cognitive behavior and marginal utility, enabling explicit control over search trajectory breadth, depth, and horizon. These improvements enable reliable extension of tool-calling beyond 20 calls when warranted. The second is Fathom-Synthesizer-4B, trained from Qwen3-4B, which converts multi-turn DeepSearch traces into structured, citation-dense DeepResearch Reports for comprehensive synthesis. Evaluated on DeepSearch benchmarks (SimpleQA, FRAMES, WebWalker, Seal0, MuSiQue) and DeepResearch-Bench, the system achieves state-of-the-art performance in the open-weights category while closely rivaling proprietary closed systems, while also demonstrating strong performance in general reasoning benchmarks: HLE, AIME-25, GPQA-Diamond, and MedQA.

Modern video-text retrieval (VTR) models excel on in-distribution benchmarks but are highly vulnerable to real-world *query shifts*, where the distribution of query data deviates from the training domain, leading to a sharp performance drop. Existing image-focused robustness solutions are inadequate to handle this vulnerability in video, as they fail to address the complex spatio-temporal dynamics inherent in these shifts. To systematically evaluate this vulnerability, we first introduce a comprehensive benchmark featuring 12 distinct types of video perturbations across five severity degrees. Analysis on this benchmark reveals that query shifts amplify the *hubness phenomenon*, where a few gallery items become dominant "hubs" that attract a disproportionate number of queries. To mitigate this, we then propose HAT-VTR (Hubness Alleviation for Test-time Video-Text Retrieval), as our baseline test-time adaptation framework designed to directly counteract hubness in VTR. It leverages two key components: a *Hubness Suppression Memory* to refine similarity scores, and *multi-granular losses* to enforce temporal feature consistency. Extensive experiments demonstrate that HAT-VTR substantially improves robustness, consistently outperforming prior methods across diverse query shift scenarios, and enhancing model reliability for real-world applications.

Diffusion Transformer has shown remarkable abilities in generating high-fidelity videos, delivering visually coherent frames and rich details over extended durations. However, existing video generation models still fall short in subject-consistent video generation due to an inherent difficulty in parsing prompts that specify complex spatial relationships, temporal logic, and interactions among multiple subjects. To address this issue, we propose BindWeave, a unified framework that handles a broad range of subject-to-video scenarios from single-subject cases to complex multi-subject scenes with heterogeneous entities. To bind complex prompt semantics to concrete visual subjects, we introduce an MLLM-DiT framework in which a pretrained multimodal large language model performs deep cross-modal reasoning to ground entities and disentangle roles, attributes, and interactions, yielding subject-aware hidden states that condition the diffusion transformer for high-fidelity subject-consistent video generation. Experiments on the OpenS2V benchmark demonstrate that our method achieves superior performance across subject consistency, naturalness, and text relevance in generated videos, outperforming existing open-source and commercial models.

Reinforcement learning (RL) has become a cornerstone for enhancing the reasoning capabilities of large language models (LLMs), with recent innovations such as Group Relative Policy Optimization (GRPO) demonstrating exceptional effectiveness. In this study, we identify a critical yet underexplored issue in RL training: low-probability tokens disproportionately influence model updates due to their large gradient magnitudes. This dominance hinders the effective learning of high-probability tokens, whose gradients are essential for LLMs' performance but are substantially suppressed. To mitigate this interference, we propose two novel methods: Advantage Reweighting and Low-Probability Token Isolation (Lopti), both of which effectively attenuate gradients from low-probability tokens while emphasizing parameter updates driven by high-probability tokens. Our approaches promote balanced updates across tokens with varying probabilities, thereby enhancing the efficiency of RL training. Experimental results demonstrate that they substantially improve the performance of GRPO-trained LLMs, achieving up to a 46.2% improvement in K&K Logic Puzzle reasoning tasks.

Online alignment (e.g., GRPO) is generally more performant than offline alignment (e.g., DPO)---but why? Drawing on prospect theory from behavioral economics, we propose a human-centric explanation. We prove that online on-policy sampling better approximates the human-perceived distribution of what the model can produce, and PPO/GRPO-style clipping---originally introduced to just stabilize training---recovers a perceptual bias in how humans perceive probability. In this sense, PPO/GRPO act as perceptual losses already. Our theory further suggests that the online/offline dichotomy is itself incidental to maximizing human utility, since we can achieve the same effect by selectively training on any data in a manner that mimics human perception, rather than restricting ourselves to online on-policy data. Doing so would allow us to post-train more quickly, cheaply, and flexibly without sacrificing performance. To this end, we propose a design pattern that explicitly incorporates perceptual distortions of probability into objectives like DPO/KTO/GRPO, creating $\textit{humanline variants}$ of them. Surprisingly, we find that these humanline variants, even when trained with offline off-policy data, can match the performance of their online counterparts on both verifiable and unverifiable tasks.

Quality-Diversity (QD) algorithms have shown remarkable success in discovering diverse, high-performing solutions, but rely heavily on hand-crafted behavioral descriptors that constrain exploration to predefined notions of diversity. Leveraging the equivalence between policies and occupancy measures, we present a theoretically grounded approach to automatically generate behavioral descriptors by embedding the occupancy measures of policies in Markov Decision Processes. Our method, AutoQD, leverages random Fourier features to approximate the Maximum Mean Discrepancy (MMD) between policy occupancy measures, creating embeddings whose distances reflect meaningful behavioral differences. A low-dimensional projection of these embeddings that captures the most behaviorally significant dimensions can then be used as behavioral descriptors for CMA-MAE, a state of the art blackbox QD method, to discover diverse policies. We prove that our embeddings converge to true MMD distances between occupancy measures as the number of sampled trajectories and embedding dimensions increase. Through experiments in multiple continuous control tasks we demonstrate AutoQD's ability in discovering diverse policies without predefined behavioral descriptors, presenting a well-motivated alternative to prior methods in unsupervised Reinforcement Learning and QD optimization. Our approach opens new possibilities for open-ended learning and automated behavior discovery in sequential decision making settings without requiring domain-specific knowledge.

Practical guidance on training Large Language Models (LLMs) to leverage Code Interpreter across diverse tasks remains lacking. We present R1-Code-Interpreter, an extension of a text-only LLM trained via multi-turn supervised fine-tuning (SFT) and reinforcement learning (RL) to autonomously generate multiple code queries during step-by-step reasoning. Unlike prior RL + tool-use efforts focused on narrow domains such as math or retrieval, we curate 144 diverse reasoning and planning tasks and show that training a general-purpose Code Interpreter across them presents significant challenges due to task heterogeneity and scarcity of effective samples. To address this, we introduce a multi-stage curriculum learning approach that partitions training samples by measured improvement potential. The RL training prioritizes samples with higher potential and gradually shifts to lower-potential ones, increasing the average RL gains from merely +3.4\% to +9.3\% across Qwen-2.5 models (3/7/14B). Our final model, R1-CI-14B, improves average accuracy on the 37 test tasks from 44.1\% to 72.4\%, outperforming text-only GPT-4o (58.6\%) and GPT-4o with Code Interpreter (70.9\%). Notably, R1-CI-14B also exhibits emergent self-checking behavior through code generation.

Temporal point processes (TPPs) are a fundamental tool for modeling event sequences in continuous time, but most existing approaches rely on autoregressive parameterizations that are limited by their sequential sampling. Recent non-autoregressive, diffusion-style models mitigate these issues by jointly interpolating between noise and data through event insertions and deletions in a discrete Markov chain. In this work, we generalize this perspective and introduce an Edit Flow process for TPPs that transports noise to data via insert, delete, and substitute edit operations. By learning the instantaneous edit rates within a continuous-time Markov chain framework, we attain a flexible and efficient model that effectively reduces the total number of necessary edit operations during generation. Empirical results demonstrate the generative flexibility of our unconditionally trained model in a wide range of unconditional and conditional generation tasks on benchmark TPPs.