ICLR 2026 Papers

Detecting individual tree crowns in tropical forests is essential to study these complex and crucial ecosystems impacted by human interventions and climate change. However, tropical crowns vary widely in size, structure, and pattern and are largely overlapping and intertwined, requiring advanced remote sensing methods applied to high-resolution imagery. Despite growing interest in tropical tree crown detection, annotated datasets remain scarce, hindering robust model development. We introduce SelvaBox, the largest open‑access dataset for tropical tree crown detection in high-resolution drone imagery. It spans three countries and contains more than $83\,000$ manually labeled crowns -- an order of magnitude larger than all previous tropical forest datasets combined. Extensive benchmarks on SelvaBox reveal two key findings: 1) higher-resolution inputs consistently boost detection accuracy; and 2) models trained exclusively on SelvaBox achieve competitive zero-shot detection performance on unseen tropical tree crown datasets, matching or exceeding competing methods. Furthermore, jointly training on SelvaBox and three other datasets at resolutions from 3 to 10 cm per pixel within a unified multi-resolution pipeline yields a detector ranking first or second across all evaluated datasets. Our dataset, code, and pre-trained weights are made public.

Large pre-trained models (LPMs) serve as universal backbones for vision and language tasks, but continual learning (CL) with frozen LPMs remains challenging, since shallow adaptation modules face the stability–plasticity dilemma and are prone to catastrophic forgetting. To address this problem, we propose MaRS (Memory-adaptive Router with Statistical control), a modular framework that decouples stable representation from adaptive capacity through three components: a frozen encoder, a slot-based memory router, and a lightweight classifier. On this basis, we design two mechanisms: (i) Statistically-Grounded Slot Expansion (SGSE) formulates expansion as a statistical decision problem, ensuring controlled growth with guarantees on false alarms and detection delay; (ii) Dual-Stage Contrastive–Distillation Adaptation (DCDA) integrates new slots through supervised contrastive learning and knowledge distillation, preserving prior knowledge without raw replay. Experiments on diverse benchmarks show that MaRS achieves state-of-the-art performance in continual learning with frozen LPMs, combining adaptability, efficiency, and retention.

Effective social intelligence simulation requires language agents to dynamically adjust reasoning depth, a capability notably absent in current studies. Existing methods either lack explicit reasoning or employ lengthy Chain-of-Thought reasoning uniformly across all scenarios, resulting in excessive token usage and inflexible social behaviors in tasks such as negotiation or collaboration. To address this, we propose an $\textbf{A}$daptive $\textbf{S}$ocial $\textbf{L}$earning ($\textbf{ASL}$) framework in this paper, aiming to improve the adaptive reasoning ability of language agents in dynamic social interactions. To this end, we first identify the hierarchical reasoning modes under such context, ranging from intuitive response to deep deliberation based on the cognitive control theory. We then develop the $\textbf{A}$daptive $\textbf{M}$ode $\textbf{P}$olicy $\textbf{O}$ptimization ($\textbf{AMPO}$) algorithm to learn the context-aware mode adaptation and reasoning. Our framework advances existing research in three key aspects: (1) Multi-granular reasoning mode design, (2) Context-aware mode switching in rich social interaction, and (3) Token-efficient reasoning with depth adaptation. Extensive experiments on the benchmark social intelligence environment verify that ASL achieves 15.6\% higher task performance than GPT-4o. Notably, our AMPO outperforms GRPO by 7.0\% with 32.8\% shorter thinking chains, demonstrating the advantages of our AMPO and the learned adaptive reasoning ability over GRPO's solution.

Probabilistic forecasting of multivariate time series is challenging due to non-stationarity, inter-variable dependencies, and distribution shifts. While recent diffusion and flow matching models have shown promise, they often ignore informative priors such as conditional means and covariances. In this work, we propose Conditionally Whitened Generative Models (CW-Gen), a framework that incorporates prior information through conditional whitening. Theoretically, we establish sufficient conditions under which replacing the traditional terminal distribution of diffusion models, namely the standard multivariate normal, with a multivariate normal distribution parameterized by estimators of the conditional mean and covariance improves sample quality. Guided by this analysis, we design a novel Joint Mean-Covariance Estimator (JMCE) that simultaneously learns the conditional mean and sliding-window covariance. Building on JMCE, we introduce Conditionally Whitened Diffusion Models (CW-Diff) and extend them to Conditionally Whitened Flow Matching (CW-Flow). Experiments on five real-world datasets with six state-of-the-art generative models demonstrate that CW-Gen consistently enhances predictive performance, capturing non-stationary dynamics and inter-variable correlations more effectively than prior-free approaches. Empirical results further demonstrate that CW-Gen can effectively mitigate the effects of distribution shift.

As Large Language Models (LLMs) are increasingly deployed in real-world applications, it is important to ensure their behaviors align with human values, societal norms, and ethical principles. However, safety alignment under Reinforcement Learning (RL) often suffers from forgetting learned general abilities, which is also known as the alignment tax. To address this issue, we introduce Null-Space constrained Policy Optimization (NSPO), a novel RL framework for LLM safety alignment while preserving their core abilities. The safety policy gradients are geometrically projected into the null space of general tasks, thereby mitigating the safety alignment tax. In addition, we theoretically prove that NSPO preserves the model's original core capabilities, while still guaranteeing a descent direction for effective safety alignment. Extensive experiments demonstrate that NSPO outperforms existing methods by a large margin, achieving state-of-the-art safety performance without sacrificing accuracy on general tasks, including math, code, and instruction-following tasks. Notably, NSPO is data-efficient and only requires 40% of public human-annotated safety data from PKU-SafeRLHF to achieve promising safety performance, without a large amount of mixed general tasks data in existing alignment methods.

Continuous-space video generation has advanced rapidly, while discrete approaches lag behind due to error accumulation and long-context inconsistency. In this work, we revisit discrete generative modeling and present Uniform discRete diffuSion with metric pAth (URSA), a simple yet powerful framework that bridges the gap with continuous approaches for the scalable video generation. At its core, URSA formulates the video generation task as an iterative global refinement of discrete spatiotemporal tokens. It integrates two key designs: a Linearized Metric Path and a Resolution-dependent Timestep Shifting mechanism. These designs enable URSA to scale efficiently to high-resolution image synthesis and long-duration video generation, while requiring significantly fewer inference steps. Additionally, we introduce an asynchronous temporal fine-tuning strategy that unifies versatile tasks within a single model, including interpolation and image-to-video generation. Extensive experiments on challenging video and image generation benchmarks demonstrate that URSA consistently outperforms existing discrete methods and achieves performance comparable to state-of-the-art continuous diffusion methods. Code and models are available at https://github.com/baaivision/URSA.

Deep neural networks often remain overconfident on Out-of-Distribution (OoD) inputs. We revisit this problem through Riemannian information geometry. We model the network's predictions as a statistical manifold and find that OoD inputs exhibit higher local Fisher-Rao sensitivity. By quantifying this sensitivity with the trace of the Fisher Information Matrix (FIM), we derive a unifying geometric connection between two common OoD signals: feature magnitude and output uncertainty. Analyzing the limitations of this multiplicative form, we extend our analysis using a product manifold construction. This provides a theoretical framework for the robust additive scores used in state-of-the-art (SOTA) detectors and motivates our final, competitive method.

Electrostatic field matching (EFM) has recently appeared as a novel physics-inspired paradigm for data generation and transfer using the idea of an electric capacitor. However, it requires modeling electrostatic fields using neural networks, which is non-trivial because of the necessity to take into account the complex field outside the capacitor plates. In this paper, we propose Interaction Field Matching (IFM), a generalization of EFM which allows using general interaction fields beyond the electrostatic one. Furthermore, inspired by strong interactions between quarks and antiquarks in physics, we design a particular interaction field realization which solves the problems which arise when modeling electrostatic fields in EFM. We show the performance on a series of toy and image data transfer problems.

Generating stylized large language model (LLM) responses via representation editing is a promising way for fine-grained output control. However, there exists an inherent trade-off: imposing a distinctive style often degrades truthfulness. Existing representation editing methods, by naively injecting style signals, overlook this collateral impact and frequently contaminate the model’s core truthfulness representations, resulting in reduced answer correctness. We term this phenomenon stylization-induced truthfulness collapse. We attribute this issue to latent coupling between style and truth directions in certain key attention heads, and propose \textbf{StyliTruth}, a mechanism that preserves stylization while keeping truthfulness intact. StyliTruth separates the style-relevant and truth-relevant subspaces in the model’s representation space via an orthogonal deflation process. This decomposition enables independent control of style and truth in their own subspaces, minimizing interference. By designing adaptive, token-level steering vectors within each subspace, we dynamically and precisely control the generation process to maintain both stylistic fidelity and truthfulness. We validate our method on multiple styles and languages. Extensive experiments and analyses show that StyliTruth significantly reduces stylization-induced truthfulness collapse and outperforms existing inference-time intervention methods in balancing style adherence with truthfulness.

Vision-based reinforcement learning can benefit from object-centric scene representation, which factorizes the visual observation into individual objects and their attributes, such as color, shape, size, and position. While such object-centric representations can extract components that generalize well for various multi-object manipulation tasks, they are prone to issues with occlusions and 3D ambiguity of object properties due to their reliance on single-view 2D image features. Furthermore, the entanglement between object configurations and camera poses complicates the object-centric disentanglement in 3D, leading to poor 3D reasoning by the agent in vision-based reinforcement learning applications. To address the lack of 3D awareness and the object-camera entanglement problem, we propose an enhanced 3D object-centric representation that utilizes multi-view 3D features and enforces more explicit 3D-aware disentanglement. The enhancement is based on the integration of the recent success of multi-view Transformer and the prototypical representation learning among the object-centric representations. The representation, therefore, can stably identify proxies of 3D positions of individual objects along with their semantic and physical properties, exhibiting excellent interpretability and controllability. Then, our proposed block transformer policy effectively performs novel tasks by assembling desired properties adaptive to the new goal states, even when provided with unseen viewpoints at test time. We demonstrate that our 3D-aware block representation is scalable to compose diverse novel scenes and enjoys superior performance in out-of-distribution tasks with multi-object manipulations under both seen and unseen viewpoints compared to existing methods.

While the advancement of large language models has spurred the development of AI agents to automate tasks, numerous use cases inherently require agents to collaborate with humans due to humans' latent preferences, domain expertise, or the need for control. To facilitate the study of human-agent collaboration, we introduce Collaborative Gym (Co-Gym), an open framework for developing and evaluating collaborative agents that engage in bidirectional communication with humans while interacting with task environments. We describe how the framework enables the implementation of new task environments and coordination between humans and agents through a flexible, non-turn-taking interaction paradigm, along with an evaluation suite that assesses both collaboration outcomes and processes. Our framework provides both a simulated condition with a reliable user simulator and a real-world condition with an interactive web application. Initial benchmark experiments across three representative tasks-creating travel plans, writing related work sections, and analyzing tabular data-demonstrate the benefits of human-agent collaboration: The best-performing collaborative agents consistently outperform their fully autonomous counterparts in task performance, achieving win rates of 86% in Travel Planning, 74% in Tabular Analysis, and 66% in Related Work when evaluated by real users. Despite these improvements, our evaluation reveals persistent limitations in current language models and agents, with communication and situational awareness failures observed in 65% and 40% of cases in the real condition, respectively. Released under the permissive MIT license, Co-Gym supports the addition of new task environments and can be used to develop collaborative agent applications, while its evaluation suite enables assessment and improvement of collaborative agents.

Modern LLM pre-training consumes vast amounts of compute and training data, making the scaling behavior, or scaling laws, of different models a key distinguishing factor. Discrete diffusion language models (DLMs) have been proposed as an alternative to autoregressive language models (ALMs). However, their scaling behavior has not yet been fully explored, with prior work suggesting that they require more data and compute to match the performance of ALMs. We study the scaling behavior of DLMs on different noise types by smoothly interpolating between masked and uniform diffusion while paying close attention to crucial hyperparameters such as batch size and learning rate. Our experiments reveal that the scaling behavior of DLMs strongly depends on the noise type and is considerably different from ALMs. While all noise types converge to similar loss values in compute-bound scaling, we find that uniform diffusion requires more parameters and less data for compute-efficient training compared to masked diffusion, making them a promising candidate in data-constrained training environments. We scale our uniform diffusion model up to 10B parameters trained for $10^{22}$ FLOPs, confirming the predicted scaling behavior and making it the largest publicly known uniform diffusion model to date. In the process of deriving the scaling laws, we reformulate the discrete diffusion ELBO in terms of signal-to-noise ratio, closing the gap to continuous diffusion theory and simplifying both theory and implementation. Training code and models are open-sourced: upon acceptance

With the rapid progress of multimodal large language models (MLLMs), AI already performs well at literature retrieval and certain reasoning tasks, serving as a capable assistant to human researchers, yet it remains far from autonomous research. The fundamental reason is that current work on scholarly paper reasoning is largely confined to a search-oriented paradigm centered on pre-specified targets, with reasoning grounded in relevance retrieval, which struggles to support researcher-style full-document understanding, reasoning, and verification. To bridge this gap, we propose ScholScan, a new benchmark for scholarly paper reasoning. ScholScan introduces a scan-oriented task setting that asks models to read and cross-check entire papers like human researchers, scanning the document to identify consistency issues. The benchmark comprises 1,800 carefully annotated questions drawn from 9 error families across 13 natural-science domains and 715 papers, and provides detailed annotations for evidence localization and reasoning traces, together with a unified evaluation protocol. We assessed 15 models across 24 input configurations and conduct a fine-grained analysis of MLLM capabilities across error families. Across the board, retrieval-augmented generation (RAG) methods yield no significant improvements, revealing systematic deficiencies of current MLLMs on scan-oriented tasks and underscoring the challenge posed by ScholScan. We expect ScholScan to be the leading and representative work of the scan-oriented task paradigm.

Developing Large Language Models (LLMs) to cooperate and compete effectively within multi-agent systems (MASs) is a critical step towards more advanced intelligence. While reinforcement learning (RL) has proven effective for enhancing reasoning in single-agent tasks, its extension to multi-turn, multi-agent scenarios remains underexplored due to the challenges of long-horizon credit assignment and agent-specific advantage estimation. To address these challenges, we introduce **MARSHAL**, an end-to-end RL framework that incentivizes **M**ulti-**A**gent **R**easoning through **S**elf-play wit**H** str**A**tegic **L**LMs in both cooperative and competitive games. MARSHAL features a turn-level advantage estimator that aligns learning signals with each interaction for credit assignment, and an agent-specific advantage normalization to stabilize multi-agent training. By learning with self-play across cooperative and competitive games, MARSHAL agents trained from Qwen3-4B develop strong strategic abilities, with up to $28.7$\% performance improvements in held-out games. More importantly, the capability acquired through self-play generalizes beyond games, yielding consistent performance gains of MASs in reasoning benchmarks. When integrated into leading MASs, our MARSHAL agent achieves significant zero-shot performance gains of up to $10.0$\% on AIME, $7.6$\% on GPQA-Diamond, and $3.5$\% on average across all benchmarks. These results establish self-play in strategic games as a powerful approach for developing generalizable multi-agent reasoning capabilities in LLMs.

Matrix operations, such as linear solves, eigendecompositions, and log determinants, are foundational building blocks for any number of downstream applications. Therefore, any broadly capable learning system should be able to effectively approximate these operations in its internal representation. Accordingly, there is great motivation to study transformers for linear algebra --- for if transformers cannot even semi-competently perform matrix operations, then we cannot expect them to form a basis for a generally intelligent system. We demonstrate that current techniques developing transformers for linear algebra have striking failure modes, prohibitive scaling, and particularly poor out-of-distribution generalization to other matrix distributions, and matrices of different sizes. Investigating further, we find that current transformer approaches operate as statistical interpolators, rather than discovering algorithms that will generalize to matrices from other distributions. Based on our understanding of these limitations, we develop a sequence of interventions that substantially improve scaling and performance, including matrix embeddings through a learnable projection, linear attention, looping, and a data pre-training distribution of structured matrices. We term the resulting method the \emph{RangeFormer}, which we show has significantly improved scaling and performance on challenging OOD matrices from the \emph{matrix market}. Moreover, with RangeFormer we show for the first time that transformers can be successfully applied to downstream tasks that involve iterative matrix operations, including Gaussian process learning, and improving the sampling distribution of randomized methods.

We present a general strategy for turning generative models into candidate solution samplers for batch Bayesian optimization (BO). The use of generative models for BO enables: large batch scaling as generative sampling, optimization of non-continuous design spaces, and high-dimensional and combinatorial design by using generative priors over feasible regions. Inspired by the success of direct preference optimization (DPO) and its variants, we show that its approach of directly training generative models using preferential rewards without the need for an intermediate reward model is extensible to the BO case. Furthermore, this framework is generalizable beyond preference-based feedback to general types of reward signals and loss functions. In essence, one can train a generative model with noisy, simple utility values directly computed from observations to then form proposal distributions whose densities are proportional to the expected utility, i.e., BO's acquisition function values. This perspective unifies recent progress in using generative models for black-box optimization and connects it with batch Bayesian optimization under a general framework. As theoretical results, we show that the generative models within the BO process approximately follow a sequence of distributions which asymptotically concentrate at the global optima under certain conditions. We also demonstrate this effect through experiments on challenging optimization problems involving large batches in high dimensions.

The unlearning capability of LLMs is vital for ensuring compliance and safety, especially when removing sensitive knowledge from deployed models. Pre-filtering methods, enabling rapid deployment without parameter changes, are a prominent unlearning approach. However, they exhibit significant robustness deficiencies against adversarial attacks: in the worst case, simple prefix attacks can induce up to a 1,150-fold surge in information leakage for fictitious entity knowledge, while composite question attacks can cause accuracy on hazardous knowledge to rebound from the 24.9% random-guess baseline to as high as 67.0%. To address this, we propose a new unlearning framework via post judgment and multi-round thinking (PoRT), which consists of three key modules. First, a data cleaning module compiles a dynamic few-shot prompt that instructs the LLM to simultaneously generate both a cleaned version of the user's query and a corresponding initial response, supported by an extensible demonstration library for adaptive defense. Second, unlike existing pre-filtering methods that typically judge based solely on prompts, our post-judgment module jointly evaluates cleaned prompts and their corresponding responses to better detect non-compliant outputs. Finally, a selective multi-round thinking process is employed to trigger LLM's self-correction for low-confidence outputs, enhancing reliability and result quality. Extensive experiments on benchmarks demonstrate PoRT's superior robustness against adversarial attacks and strong unlearning effectiveness without compromising general model utility. Code is available at https://github.com/ChnIRuI/PoRT_LLM_Unlearning

Collecting web data to train deep models has become increasingly common, raising concerns about unauthorized data usage. To mitigate this issue, unlearnable examples introduce imperceptible perturbations into data, preventing models from learning effectively. However, existing methods typically rely on deep neural networks as surrogate models for perturbation generation, resulting in significant computational costs. In this work, we propose Perturbation-Induced Linearization (PIL), a computationally efficient yet effective method that generates perturbations using only linear surrogate models. PIL achieves comparable or better performance than existing surrogate-based methods while reducing computational time dramatically. We further reveal a key mechanism underlying unlearnable examples: inducing linearization to deep models, which explains why PIL can achieve competitive results in a very short time. Beyond this, we provide an analysis about the property of unlearnable examples under percentage-based partial perturbation. Our work not only provides a practical approach for data protection but also offers insights into what makes unlearnable examples effective.

Hard negative generation (HNG) provides valuable signals for deep learning, but existing methods mostly rely on local correlations while neglecting the global geometry of the embedding space. This limitation often leads to weak discrimination, particularly in cross-modal hashing, which obtains compact binary codes. We propose Deep Global-sense Hard-negative Discriminative Generation Hashing (DGHDGH), a framework that constructs a structured graph with dual-iterative message propagation to capture global correlations, and then performs difficulty-adaptive, channel-wise interpolation to synthesize semantically consistent hard negatives aligned with global Hamming geometry. Our approach yields more informative negatives, sharpens semantic boundaries in the Hamming co-space, and substantially enhances cross-modal retrieval. Experiments on multiple benchmarks consistently demonstrate improvements in retrieval accuracy, verifying the discriminative advantages brought by global-sense HNG in cross-modal hashing.