ICLR 2026 Papers

The rapid rise of image generation calls for detection methods that are both interpretable and reliable. Existing approaches, though accurate, act as black boxes and fail to generalize to out-of-distribution data, while multi-modal large language models (MLLMs) provide reasoning ability but often hallucinate. To address these issues, we construct \textbf{FakeXplained} dataset of AI-generated images annotated with bounding boxes and descriptive captions that highlight synthesis artifacts, forming the basis for human-aligned, visually grounded reasoning. Leveraging \textbf{FakeXplained}, we develop \textbf{FakeXplainer} which fine-tunes MLLMs with a progressive training pipeline, enabling accurate detection, artifact localization, and coherent textual explanations. Extensive experiments show that \textbf{FakeXplainer} not only sets a new state-of-the-art in detection and localization accuracy ($98.2\%$ accuracy, $36.0\%$ IoU), but also demonstrates strong robustness and out-of-distribution generalization, uniquely delivering spatially grounded, human-aligned rationales. The code and dataset are available at: \href{https://github.com/Gennadiyev/FakeXplain}{https://github.com/Gennadiyev/FakeXplain}.

Sparse Autoencoders (SAEs) are a powerful dictionary learning technique for decomposing neural network activations, translating the hidden state into human ideas with high semantic value despite no external intervention or guidance. However, this technique has rarely been applied outside of the textual domain, limiting theoretical explorations of feature decomposition. We present the first application of SAEs to the 3D domain, analyzing the features used by a state-of-the-art 3D reconstruction VAE applied to 53k 3D models from the Objaverse dataset. We observe that the network encodes discrete rather than continuous features, leading to our key finding: such models approximate a discrete state space, driven by phase-like transitions from feature activations. Through this state transition framework, we address three otherwise unintuitive behaviors — the inclination of the reconstruction model towards positional encoding representations, the sigmoidal behavior of reconstruction loss from feature ablation, and the bimodality in the distribution of phase transition points. This final observation suggests the model actively manipulates the interference caused by superposition to prioritize the saliency of different features. Our work not only compiles and explains unexpected phenomena regarding feature decomposition, but also describes a generally applicable, state-based feature framework.

The generalization of learning-based high dynamic range (HDR) fusion is often limited by the availability of training data, as collecting large-scale HDR images from dynamic scenes is both costly and technically challenging. To address these challenges, we propose S2R-HDR, the first large-scale high-quality synthetic dataset for HDR fusion, with 24,000 HDR samples. Using Unreal Engine 5, we design a diverse set of realistic HDR scenes that encompass various dynamic elements, motion types, high dynamic range scenes, and lighting. Additionally, we develop an efficient rendering pipeline to generate realistic HDR images. To further mitigate the domain gap between synthetic and real-world data, we introduce S2R-Adapter, a domain adaptation designed to bridge this gap and enhance the generalization ability of models. Experimental results on real-world datasets demonstrate that our approach achieves state-of-the-art HDR fusion performance.

Independence testing involves determining whether two variables are independent based on observed samples, which is a fundamental problem in statistics and machine learning. Existing testing methods, such as HSIC, can theoretically detect broad forms of dependence, but may sacrifice statistical power when applied to limited samples with background knowledge of the distribution. In this paper, we focus on the linear non-Gaussian data, a widely supported model in scientific data analysis and causal discovery, where variables are linked linearly with noise terms that are non-Gaussian distributed. We provide a new theoretical characterization of independence in this case, showing that constancy of the conditional mean and variance is sufficient to guarantee independence under linear non-Gaussian models. Building on this result, we develop a kernel-based testing framework with provable asymptotic guarantees. Extensive experiments on synthetic and real-world datasets demonstrate that our method achieves higher power than existing approaches and significantly improves downstream causal discovery performance.

Latent diffusion has achieved remarkable success in image generation, with high sampling efficiency. However, this framework might suffer from posterior collapse when applied to time series. In this work, we first show that latent diffusion with a collapsed posterior degenerates into a much weaker generative model: variational autoencoder (VAE). This finding highlights the significance of addressing the problem. We then introduce a principled method: dependency measures, which quantify the sensitivity of a recurrent decoder to input variables. Through this method, we confirm that posterior collapse seriously affects latent time-series diffusion on real time series. For example, the latent variable has an exponentially decreasing impact on the decoder over time. Building on our theoretical and empirical studies, we finally introduce a new framework: posterior-stable latent diffusion, which interprets the diffusion process as a type of variational inference. In this way, it eliminates the use of risky KL regularization and penalizes decoder insensitivity. Extensive experiments on multiple real time-series datasets show that our new framework is with a highly stable posterior and notably outperforms previous baselines in time series synthesis.

DeepSeek-R1-Zero has successfully demonstrated the emergence of reasoning capabilities in LLMs purely through Reinforcement Learning (RL). Inspired by this breakthrough, we explore how RL can be utilized to enhance the reasoning capability of MLLMs. However, direct training with RL struggles to activate complex reasoning capabilities such as questioning and reflection in MLLMs, due to the absence of substantial high-quality multimodal reasoning data. To address this issue, we propose the reasoning MLLM, Vision-R1, to improve multimodal reasoning capability. Specifically, we first construct a high-quality multimodal CoT dataset without human annotations by leveraging an existing MLLM and DeepSeek-R1 through modality bridging and data filtering to obtain a 200K multimodal CoT dataset, Vision-R1-cold dataset. It serves as cold-start initialization data for Vision-R1. To mitigate the optimization challenges caused by overthinking after cold start, we propose Progressive Thinking Suppression Training (PTST) strategy and employ Group Relative Policy Optimization (GRPO) with the hard formatting result reward function to gradually refine the model's ability to learn correct and complex reasoning processes on the multimodal math dataset. Comprehensive experiments show our model achieves an average improvement of $\sim$6\% across various multimodal math reasoning benchmarks using only a 10K multimodal math data during RL training. Vision-R1-7B achieves a 73.5\% accuracy on the widely used MathVista benchmark, which is only 0.4\% lower than the leading reasoning model, OpenAI O1. Scaling up the amount of multimodal math data in the RL training, Vision-R1-32B and Vison-R1-72B achieves 76.4\% and 78.2\% MathVista benchmark scores, respectively.

Few-shot text classification is a critical real-world task for which Large Language Models (LLMs) have shown great promise. However, their high inference costs and lack of interpretability limit their practical use. While Concept Bottleneck Models (CBMs) offer an efficient and interpretable alternative, their reliance on training surrogate models makes them incompatible with few-shot scenarios. To bridge this gap, we introduce a novel CBM paradigm that relies solely on sample-concept similarity to make predictions. We ensure the effectiveness of our concepts through a prototypical-discriminative dual-level architecture and a dynamic concept refinement mechanism. Extensive experiments show that with as few as 10 training samples, our method surpasses prior CBMs and even achieves performance comparable to LLMs. The code is available at https://anonymous.4open.science/r/StructCBM-EB1E.

We present \textbf{Frequency-Aware Dynamic Gaussian Splatting (FAGS)}, a novel approach to mitigating motion blur in 4D reconstruction, particularly under novel viewpoints. This blur stems from a fundamental spectral conflict in existing methods, which struggle to \textbf{balance high-frequency rendering details with high-frequency motion.} FAGS addresses this challenge with two key innovations. First, we introduce a frequency-differentiated Gaussian kernel that refines the alpha-blending process of 3D Gaussian Splatting. By adaptively classifying Gaussians into two types—a slowly varying kernel for smooth, low-frequency regions and a sharp-transitioning kernel for high-frequency boundaries—our method explicitly separates representation responsibilities, preserving fine details without sacrificing continuity. Second, we propose a Fourier-Deformation Network that enhances motion expressiveness. This network employs high-frequency Fourier embeddings to capture diverse motion patterns by learning amplitudes across frequency components. To further improve accuracy, we integrate a frequency-aware gate in fusion module, which predicts and regulates the relative deformation of each Gaussian. Extensive experiments on both synthetic and real-world 4D benchmarks demonstrate that FAGS significantly reduces motion blur and enhances structural details, achieving state-of-the-art performance.

In this paper, we uniquely study the adversarial robustness of deep neural networks (NN) for classification tasks against that of optimal classifiers. We look at the smallest magnitude of possible additive perturbations that can change a classifier's output. We provide a matrix-theoretic explanation of the adversarial fragility of deep neural networks for classification. In particular, our theoretical results show that a neural network's adversarial robustness can degrade as the input dimension $d$ increases. Analytically, we show that neural networks' adversarial robustness can be only $1/\sqrt{d}$ of the best possible adversarial robustness of optimal classifiers. Our theories match remarkably well with numerical experiments of practically trained NN, including NN for ImageNet images. The matrix-theoretic explanation is consistent with an earlier information-theoretic feature-compression-based explanation for the adversarial fragility of neural networks.

Multimodal retrieval still leans on embedding-based models like CLIP for fast vector search over pre-computed image embeddings. Yet, unlike text retrieval where joint-encoder rerankers are standard, comparable vision–language rerankers are largely absent. We find that seminal joint encoders such as BLIP are severely bottlenecked by an expensive visual feature-extraction stage, preventing practical deployment at scale. Motivated by this bottleneck, we introduce EDJE , an Efficient Discriminative Joint Encoder that precomputes vision tokens offline and compresses them via a lightweight attention-based adapter, so online inference runs only a compact joint encoder over a small set of visual tokens plus the text. EDJE preserves strong retrieval performance while drastically reducing storage and online compute, enabling high-throughput inference. Specifically, EDJE processes 50k image–text pairs/second while requiring 49kB of disk storage per image, matching prior art on Flickr (zero-shot) and COCO (fine-tuned) retrieval.

Growing concerns over the theft and misuse of Large Language Models (LLMs) underscore the need for effective fingerprinting to link a model to its original version and detect misuse. We define five essential properties for a successful fingerprint: Transparency, Efficiency, Persistence, Robustness, and Unforgeability. We present a novel fingerprinting framework that provides verifiable proof of ownership while preserving fingerprint integrity. Our approach makes two main contributions. First, a "chain and hash" technique that cryptographically binds fingerprint prompts to their responses, preventing collisions and enabling irrefutable ownership claims. Second, we address a realistic threat model in which instruction-tuned models' output distribution can be significantly altered through meta-prompts. By incorporating random padding and varied meta-prompt configurations during training, our method maintains robustness even under significant output style changes. Experiments show that our framework securely proves ownership, resists both benign transformations (e.g., fine-tuning) and adversarial fingerprint removal, and extends to fingerprinting LoRA adapters.

Nearest Neighbor Search (NNS) is a fundamental problem in data structures with wide-ranging applications, such as web search, recommendation systems, and, more recently, retrieval-augmented generations (RAG). In such recent applications, in addition to the relevance (similarity) of the returned neighbors, diversity among the neighbors is a central requirement. In this paper, we develop principled welfare-based formulations in NNS for realizing diversity across attributes. Our formulations are based on welfare functions---from mathematical economics---that satisfy central diversity (fairness) and relevance (economic efficiency) axioms. With a particular focus on Nash social welfare, we note that our welfare-based formulations provide objective functions that adaptively balance relevance and diversity in a query-dependent manner. Notably, such a balance was not present in the prior constraint-based approach, which forced a fixed level of diversity and optimized for relevance. In addition, our formulation provides a parametric way to control the trade-off between relevance and diversity, providing practitioners with flexibility to tailor search results to task-specific requirements. We develop efficient nearest neighbor algorithms with provable guarantees for the welfare-based objectives. Notably, our algorithm can be applied on top of any standard ANN method (i.e., use standard ANN method as a subroutine) to efficiently find neighbors that approximately maximize our welfare-based objectives. Experimental results demonstrate that our approach is practical and substantially improves diversity while maintaining high relevance of the retrieved neighbors.

Vision-language-action (VLA) models have shown promise as generalist robotic policies by jointly leveraging visual, linguistic, and proprioceptive modalities to generate action trajectories. While recent benchmarks have advanced VLA research in domestic tasks, professional science-oriented domains remain underexplored. We introduce AutoBio, a simulation framework and benchmark designed to evaluate robotic automation in biology laboratory environments—an application domain that combines structured protocols with demanding precision and multimodal interaction. AutoBio extends existing simulation capabilities through a pipeline for digitizing real-world laboratory instruments, specialized physics plugins for mechanisms ubiquitous in laboratory workflows, and a rendering stack that support dynamic instrument interfaces and transparent materials through physically based rendering. Our benchmark comprises biologically grounded tasks spanning three difficulty levels, enabling standardized evaluation of language-guided robotic manipulation in experimental protocols. We provide infrastructure for demonstration generation and seamless integration with VLA models. Baseline evaluations with SOTA VLA models reveal significant gaps in precision manipulation, visual reasoning, and instruction following in scientific workflows. By releasing AutoBio, we aim to catalyze research on generalist robotic systems for complex, high-precision, and multimodal professional environments.

Vision-language alignment is crucial for various downstream tasks such as cross-modal generation and retrieval. Previous multimodal approaches like CLIP utilize InfoNCE to maximize mutual information, primarily aligning pairwise samples across modalities while overlooking distributional differences. In addition, InfoNCE has inherent conflict in terms of alignment and uniformity in multimodality, leading to suboptimal alignment with modality gaps. To overcome the limitations, we propose CS-Aligner, a novel framework that performs distributional vision-language alignment by integrating Cauchy-Schwarz (CS) divergence with mutual information. CS-Aligner captures both the global distribution information of each modality and the pairwise semantic relationships. We find that the CS divergence seamlessly addresses the InfoNCE's alignment-uniformity conflict and serves complementary roles with InfoNCE, yielding tighter and more precise alignment. Moreover, by introducing distributional alignment, CS-Aligner enables incorporating additional information from unpaired data and token-level representations, enhancing flexible and fine-grained alignment in practice. Experiments on text-to-image generation and cross-modality retrieval tasks demonstrate the effectiveness of our method on vision-language alignment.

Sparse autoencoders (SAEs) are widely used to extract sparse, interpretable latents from transformer activations. We test whether commonly used SAE quality metrics and automatic explanation pipelines can distinguish trained transformers from randomly initialized ones (e.g., where parameters are sampled i.i.d. from a Gaussian). Over a wide range of Pythia model sizes and multiple randomization schemes, we find that, in many settings, SAEs trained on randomly initialized transformers produce auto-interpretability scores and reconstruction metrics that are similar to those from trained models. These results show that high aggregate auto-interpretability scores do not, by themselves, guarantee that learned, computationally relevant features have been recovered. We therefore recommend treating common SAE metrics as useful but insufficient proxies for mechanistic interpretability and argue for routine randomized baselines and targeted measures of feature 'abstractness'.

Despite achieving significant progress in 2D image understanding, large multimodal models (LMMs) struggle in the physical world due to the lack of spatial representation. Typically, existing 3D LMMs mainly embed 3D positions as fixed spatial prompts within visual features to represent the scene. However, these methods are limited to understanding the static background and fail to capture temporally varying dynamic objects. In this paper, we propose LLaVA-4D, a general LMM framework with a novel spatiotemporal prompt for visual representation in 4D scene understanding. The spatiotemporal prompt is generated by encoding 3D position and 1D time into a dynamic-aware 4D coordinate embedding. Moreover, we demonstrate that spatial and temporal components disentangled from visual features are more effective in distinguishing the background from objects. This motivates embedding the 4D spatiotemporal prompt into these features to enhance the dynamic scene representation. By aligning visual spatiotemporal embeddings with language embeddings, LMMs gain the ability to understand both spatial and temporal characteristics of static background and dynamic objects in the physical world. Additionally, we construct a 4D vision-language dataset with spatiotemporal coordinate annotations for instruction fine-tuning LMMs. Extensive experiments have been conducted to demonstrate the superiority of our method on various tasks of 4D scene understanding. Our code will be open-sourced on paper acceptance.

Diffusion Transformers (DiTs) offer state-of-the-art fidelity in image and video synthesis, but their iterative sampling process remains a major bottleneck due to the high cost of transformer forward passes at each timestep. To mitigate this, feature caching has emerged as a training-free acceleration technique that reuses or forecasts hidden representations. However, existing methods often apply a uniform caching strategy across all feature dimensions, ignoring their heterogeneous dynamic behaviors. Therefore, we adopt a new perspective by modeling hidden feature evolution as a mixture of ODEs across dimensions, and introduce \textbf{HyCa}, a Hybrid ODE solver inspired caching framework that applies dimension-wise caching strategies. HyCa achieves near-lossless acceleration across diverse domains and models, including 5.56$\times$ speedup on FLUX and HunyuanVideo, 6.24$\times$ speedup on Qwen-Image and Qwen-Image-Edit without retraining. \emph{Our code is in supplementary material and will be released on Github.}

Diffusion large language models (dLLMs) enable any-order generation, but this flexibility enlarges the attack surface: harmful spans may appear at arbitrary positions, and template-based prefilling attacks such as DIJA bypass response-level refusals. We introduce A2D (Any-Order, Any-Step Defense), a token-level alignment method that aligns dLLMs to emit an [EOS] refusal signal whenever harmful content arises. By aligning safety directly at the token-level under randomized masking, A2D achieves robustness to both any-decoding-order and any-step prefilling attacks under various conditions. It also enables real-time monitoring: dLLMs may begin a response but automatically terminate if unsafe continuation emerges. On safety benchmarks, A2D consistently prevents the generation of harmful outputs, slashing DIJA success rates from over 80\% to near-zero (1.3\% on LLaDA-8B-Instruct, 0.0\% on Dream-v0-Instruct-7B), and thresholded [EOS] probabilities allow early rejection, yielding up to 19.3× faster safe termination.

Existing video-to-audio (V2A) generation methods predominantly rely on text prompts alongside visual information to synthesize audio. However, two critical bottlenecks persist: semantic granularity gaps in training data (e.g., conflating acoustically distinct sounds like different dog barks under coarse labels), and textual ambiguity in describing microacoustic features (e.g., "metallic clang" failing to distinguish impact transients and resonance decay). These bottlenecks make it difficult to perform fine-grained sound synthesis using text-controlled modes. To address these limitations, we propose **AC-Foley**, an audio-conditioned V2A model that directly leverages reference audio to achieve precise and fine-grained control over generated sounds. This approach enables: fine-grained sound synthesis (e.g., footsteps with distinct timbres on wood, marble, or gravel), timbre transfer (e.g., transforming a violin’s melody into the bright, piercing tone of a suona), zero-shot generation of sounds (e.g., creating unique weapon sound effects without training on firearm datasets) and better audio quality. By directly conditioning on audio signals, our approach bypasses the semantic ambiguities of text descriptions while enabling precise manipulation of acoustic attributes. Empirically, AC-Foley achieves state-of-the-art performance for Foley generation when conditioned on reference audio, while remaining competitive with SOTA video-to-audio methods even without audio conditioning.