ICLR 2026 论文列表

Reconstructing 3D scenes from sparse images remains a challenging task due to the difficulty of recovering accurate geometry and texture without optimization. Recent approaches leverage generalizable models to generate 3D scenes using 3D Gaussian Splatting (3DGS) primitive. However, they often fail to produce continuous surfaces and instead yield discrete, color-biased point clouds that appear plausible at normal resolution but reveal severe artifacts under close-up views. To address this issue, we present SurfSplat, a feedforward framework based on 2D Gaussian Splatting (2DGS) primitive, which provides stronger anisotropy and higher geometric precision. By incorporating a surface continuity prior and a forced alpha blending strategy, SurfSplat reconstructs coherent geometry together with faithful textures. Furthermore, we introduce High-Resolution Rendering Consistency (HRRC), a new evaluation metric designed to evaluate high-resolution reconstruction quality. Extensive experiments on RealEstate10K, DL3DV, and ScanNet demonstrate that SurfSplat consistently outperforms prior methods on both standard metrics and HRRC, establishing a robust solution for high-fidelity 3D reconstruction from sparse inputs.

Biomolecular interactions play a critical role in biological processes. While recent breakthroughs like AlphaFold 3 have enabled accurate modeling of biomolecular complex structures, predicting binding affinity remains challenging mainly due to limited high-quality data. Recent methods are often specialized for specific types of biomolecular interactions, limiting their generalizability. In this work, we repurpose AlphaFold 3 for representation learning to predict binding affinity, a non-trivial task that requires shifting from generative structure prediction to encoding observed geometry, simplifying the heavily conditioned trunk module, and designing a framework to jointly capture sequence and structural information. To address these challenges, we introduce the **Atom-level Diffusion Transformer (ADiT)**, which takes sequence and structure as inputs, employs a unified tokenization scheme, integrates diffusion transformers, and removes dependencies on multiple sequence alignments and templates. We pre-train three ADiT variants on the PDB dataset with a denoising objective and evaluate them across protein-ligand, drug-target, protein-protein, and antibody-antigen interactions. The model achieves state-of-the-art or competitive performance across benchmarks, scales effectively with model size, and successfully identifies wet-lab validated affinity-enhancing antibody mutations, establishing a generalizable framework for biomolecular interactions. We plan to release the code upon acceptance.

Large Language Model (LLM) agents show great promise for complex multi-turn tool-use tasks, but their development is often hampered by the extreme scarcity of high-quality training data. Supervised fine-tuning (SFT) on synthetic data leads to overfitting, whereas standard reinforcement learning (RL) struggles with a critical cold-start problem and training instability. To address these challenges, we introduce $\textbf{Environment Tuning}$, a novel training paradigm that enables agents to learn complex behaviors directly from problem instances without relying on pre-collected expert trajectories. $\textbf{Environment Tuning}$ orchestrates this learning process through a structured curriculum, actionable environment augmentation that provides corrective feedback, and fine-grained progress rewards to ensure stable and efficient exploration. Using only 400 problem instances from Berkeley Function-Calling Leaderboard (BFCL) benchmark, our method not only achieves competitive in-distribution performance against strong baselines but also demonstrates superior out-of-distribution generalization, overcoming the performance collapse common to SFT-based approaches. Our work presents a paradigm shift from supervised fine-tuning on static trajectories to dynamic, environment-based exploration, paving the way for training more robust and data-efficient agents.

The widespread deployment of LLM-based agents is likely to introduce a critical privacy threat: malicious agents that proactively engage others in multi-turn interactions to extract sensitive information. However, the evolving nature of such dynamic dialogues makes it challenging to anticipate emerging vulnerabilities and design effective defenses. To tackle this problem, we present a search-based framework that alternates between improving attack and defense strategies through the simulation of privacy-critical agent interactions. Specifically, we employ LLMs as optimizers to analyze simulation trajectories and iteratively propose new agent instructions. To explore the strategy space more efficiently, we further utilize parallel search with multiple threads and cross-thread propagation. Through this process, we find that attack strategies escalate from direct requests to sophisticated tactics, such as impersonation and consent forgery, while defenses evolve from simple rule-based constraints to robust identity-verification state machines. The discovered attacks and defenses transfer across diverse scenarios and backbone models, demonstrating strong practical utility for building privacy-aware agents.

Recent inversion-free, flow-based image editing methods such as FlowEdit leverages a pre-trained noise-to-image flow model such as Stable Diffusion 3, enabling text-driven manipulation by solving an ordinary differential equation (ODE). While the lack of exact latent inversion is a core advantage of these methods, it often results in unstable editing trajectories and poor source consistency. To address this limitation, we propose {\em FlowAlign}, a novel inversion-free flow-based framework for consistent image editing with principled trajectory control. FlowAlign introduces a flow-matching loss as a regularization mechanism to promote smoother and more stable trajectories during the editing process. Notably, the flow-matching loss is shown to explicitly balance semantic alignment with the edit prompt and structural consistency with the source image along the trajectory. Furthermore, FlowAlign naturally supports reverse editing by simply reversing the ODE trajectory, highliting the reversible and consistent nature of the transformation. Extensive experiments demonstrate that FlowAlign outperforms existing methods in both source preservation and editing controllability.

ViTs are general and accurate, and address many tasks, but ViTs are slow, and are not always practical when efficiency is key. Existing methods for faster ViTs design hybrid non-ViT architectures, losing generality, or shrink their tokens, sacrificing accuracy. While many non-ViT architectures are both fast and accurate, they cannot flexibly process other input shapes, pre-train by SOTA self-supervised learning, reduce computation by dropping tokens, and more like ViTs can. We make ViTs faster by reducing patch token width while increasing global token width by adding a new Jumbo token. Our wider Jumbo token is processed by its own wider FFN to increase model capacity. Yet our Jumbo FFN is efficient: it processes a single token, for speed, and its parameters are shared across all layers, for memory. Crucially, our Jumbo is attention-only and non-hierarchical, like a plain ViT, so it is simple, scalable, flexible, and compatible with ViT methods new and old. Jumbo improves over ViT baselines with Registers from Nano to Large scales while maintaining speed/throughput on ImageNet-1K (↑0.1−13%). Jumbo also improves MAE pre-training (↑4.9% linear probing on ImageNet-1K), test-time adaptation (↑5.2% on ImageNet-C), and time series modeling. Our Jumbo models even achieve better speed-accuracy trade-offs than specialized non-ViT compute-efficient models, while maintaining plain-ViT compatibility for practicality.

We study the problem of minimizing polarization and disagreement in the Friedkin–Johnsen opinion dynamics model under incomplete information. Unlike prior work that assumes a static setting with full knowledge of users' innate opinions, we address the more realistic online setting where innate opinions are unknown and must be learned through sequential observations. This novel setting, which naturally mirrors periodic interventions on social media platforms, is formulated as a regret minimization problem, establishing a key connection between algorithmic interventions on social media platforms and theory of multi-armed bandits. In our formulation, a learner observes only a scalar feedback of the overall polarization and disagreement after an intervention. For this novel bandit problem, we propose a two-stage algorithm based on low-rank matrix bandits. The algorithm first performs subspace estimation to identify an underlying low-dimensional structure, and then employs a linear bandit algorithm within the compact dimensional representation derived from the estimated subspace. We prove that our algorithm achieves an $\widetilde{O}(\sqrt{T})$ cumulative regret over any time horizon $T$. Empirical results validate that our algorithm significantly outperforms a linear bandit baseline in terms of both cumulative regret and running time.

Robust 3D hand reconstruction is challenging in egocentric vision due to depth ambiguity, self-occlusion, and complex hand-object interactions. Prior works attempt to mitigate the challenges by scaling up training data or incorporating auxiliary cues, often falling short of effectively handling unseen contexts. In this paper, we introduce EgoHandICL, the first in-context learning (ICL) framework for 3D hand reconstruction that achieves strong semantic alignment, visual consistency, and robustness under challenging egocentric conditions. Specifically, we develop (i) complementary exemplar retrieval strategies guided by vision–language models (VLMs), (ii) an ICL-tailored tokenizer that integrates multimodal context, and (iii) a Masked Autoencoders (MAE)-based architecture trained with 3D hand–guided geometric and perceptual objectives. By conducting comprehensive experiments on the ARCTIC and EgoExo4D benchmarks, our EgoHandICL consistently demonstrates significant improvements over state-of-the-art 3D hand reconstruction methods. We further show EgoHandICL’s applicability by testing it on real-world egocentric cases and integrating it with EgoVLMs to enhance their hand–object interaction reasoning. Our code and data will be publicly available.

Monocular 3D Gaussian Splatting SLAM suffers from critical limitations in time efficiency, geometric accuracy, and multi-view consistency. These issues stem from the time-consuming $\textit{Train-from-Scratch}$ optimization and the lack of inter-frame scale consistency from single-frame geometry priors. We contend that a feed-forward paradigm, leveraging multi-frame context to predict Gaussian attributes directly, is crucial for addressing these challenges. We present Flash-Mono, a system composed of three core modules: a feed-forward prediction frontend, a 2D Gaussian Splatting mapping backend, and an efficient hidden-state-based loop closure module. We trained a recurrent feed-forward frontend model that progressively aggregates multi-frame visual features into a hidden state via cross attention and jointly predicts camera poses and per-pixel Gaussian properties. By directly predicting Gaussian attributes, our method bypasses the burdensome per-frame optimization required in optimization-based GS-SLAM, achieving a $\textbf{10x}$ speedup while ensuring high-quality rendering. The power of our recurrent architecture extends beyond efficient prediction. The hidden states act as compact submap descriptors, facilitating efficient loop closure and global $\mathrm{Sim}(3)$ optimization to mitigate the long-standing challenge of drift. For enhanced geometric fidelity, we replace conventional 3D Gaussian ellipsoids with 2D Gaussian surfels. Extensive experiments demonstrate that Flash-Mono achieves state-of-the-art performance in both tracking and mapping quality, highlighting its potential for embodied perception and real-time reconstruction applications.

We propose a method for audio-visual knowledge distillation. Existing methods typically distill from the latent embeddings or outputs. The former requires matching feature dimensions, if not the same architecture, between teacher and student models while the latter supports any teacher-student pairing, but tends to be less performant. Unlike them, we do not explicitly distill from the latent embeddings or outputs, but the pairwise relationships between embeddings across samples for each modality; this is realized as a kernel, which is the crux of our method, ``Kernelized Token Distillation (KTD)''. Specifically, we tokenize and embed the input for a given modality, and compute the Gram matrix across tokens, from which we distill. As audio and visual modalities afford different information for a task, we adaptively modulate distillation by measuring the entropy of each modality, leading to an Entropty-Monitored Kernelized Token Distillation (EM-KTD) scheme. Our method allows flexibility in complexity of kernel function to model relationships across tokens, which are selectively distilled to ensure high-fidelity supervision for the student. We evaluate EM-KTD on VGGSound and AVS-Bench, where we use 94\% fewer parameters than the teacher while preserving 96.9\% in performance for audio-visual event recognition and 96.5\% on audio-visual segmentation.

While Large Vision-Language Models (LVLMs) have achieved substantial progress in video understanding, their application to long video reasoning is hindered by uniform frame sampling and static textual reasoning, which are inefficient and struggle to handle visually intensive video tasks. To overcome these challenges, in this paper, we introduce the concept of thinking with long videos and propose a novel framework FrameThinker. Within this framework, LVLMs are able to iteratively interrogate video content. Developing such video reasoning capabilities in LVLMs presents notable challenges, particularly in adapting the model to new video actions (e.g. select frame), and designing reward functions to guide LVLMs to adopt the newly introduced action. To solve these challenges, we propose a two-phase training strategy, first employing Supervised Fine-Tuning (SFT) to instill fundamental action capabilities, followed by Reinforcement Learning (RL) to optimize a strategic decision-making policy. Notably, in this RL phase, we conduct an in-depth and comprehensive exploration of the reward design for each action and format reward. Extensive experiments on reasoning benchmarks like Video-Holmes, LongVideo-Reason, and long-video understanding benchmarks such as LongVideoBench, MLVU, VideoMME, and LVBench, demonstrate that FrameThinker gets a significant average improvement of +10.4\% over baselines while drastically reducing the number of processed frames. Most notably, our 7B model, FrameThinker establishes a new state-of-the-art on LongVideo-Reason, achieving 76.1\% accuracy using an average of only 20.6 frames. This not only outperforms the competitive LongVILA-R1 (72.0\%) but does so with over 20x fewer frames (vs. 512), demonstrating unparalleled efficiency and effectiveness. Our code is available at: \url{https://anonymous.4open.science/r/FrameThinker-B5FD}.

Computed Tomography (CT) is one of the most widely used and diagnostically information-dense imaging modalities, covering critical organs such as the heart, lungs, liver, and colon. Clinical interpretation relies on both \textbf{slice-driven} local features (e.g., sub-centimeter nodules, lesion boundaries) and \textbf{volume-driven} spatial representations (e.g., tumor infiltration, inter-organ anatomical relations). However, existing Large Vision–Language Models (LVLMs) remain fragmented in CT slice versus volumetric understanding: slice-driven LVLMs show strong generalization but lack cross-slice spatial consistency, while volume-driven LVLMs explicitly capture volumetric semantics but suffer from coarse granularity and poor compatibility with slice inputs. The absence of a unified modeling paradigm constitutes a major bottleneck for the clinical translation of medical LVLMs. We present \textbf{OmniCT}, a powerful unified slice–volume LVLM for CT scans, which makes three contributions: \textbf{(i) Spatial Consistency Enhancement (SCE):} volumetric slice composition combined with tri-axial positional encoding introduces volumetric consistency, and an MoE hybird projection enables efficient slice–volume adaptation; \textbf{(ii) Organ-level Semantic Enhancement (OSE):} segmentation and ROI localization explicitly align anatomical regions, emphasizing lesion- and organ-level semantics; \textbf{(iii) MedEval-CT:} the largest slice–volume CT dataset and hybrid benchmark integrates multi-level metrics for unified evaluation. OmniCT consistently outperforms existing methods with a substantial margin across diverse clinical tasks, satisfies both micro-level detail sensitivity and macro-level spatial reasoning, and establishes a new paradigm for cross-dimensional medical imaging modeling. Our project is available at \href{https://anonymous.4open.science/r/OmniCT}{link}.

Deep spiking neural networks (SNNs) hold immense promise for low-power event-driven computing, but their direct training via backpropagation through time (BPTT) incurs prohibitive memory cost, which limits their scalability. Existing memory-saving approaches, such as online learning, BPTT-to-BP, and reversible networks, compromise accuracy, training speed, or applicability. In this work, we propose a novel and broadly applicable pipeline for memory-efficient SNN training that preserves BPTT's accuracy. Our pipeline integrates layer-wise gradient checkpointing with lossless spike compression to eliminate internal state storage and reduce the memory cost of per-layer input spikes. We also introduce a multi-stage checkpoint adjustment strategy that adaptively refines checkpoint placement based on profiling results to further optimize memory usage and improve training speed. Wrapped in an optimization pass, the pipeline automatically restructures the computation flow before training with minimal user effort. Extensive experiments on diverse architectures and tasks demonstrate up to $8\times$ memory efficiency gains with $\le 20\\%$ speed reduction and no accuracy loss. Our method provides a practical solution for efficient and scalable SNN training. Code will be available upon acceptance.

Existing rule-based explanations for Graph Neural Networks (GNNs) provide global interpretability but often optimize and assess fidelity in an intermediate, uninterpretable concept space, overlooking the grounding quality of the final subgraph explanations for end users. This gap yields explanations that may appear faithful yet be unreliable in practice. To this end, we propose LogicXGNN, a post hoc framework that constructs logical rules over reliable predicates explicitly designed to capture the GNN's message-passing structure, thereby ensuring effective grounding. We further introduce data-grounded fidelity ($Fid_D$), a realistic metric that evaluates explanations in their final-graph form, along with complementary utility metrics such as coverage and validity. Across extensive experiments, LogicXGNN improves $Fid_D$ by over 20% on average relative to state-of-the-art methods while being 10-100 times faster. With strong scalability and utility performance, LogicXGNN produces explanations that are faithful to the model's logic and reliably grounded in observable data.

The recent paradigm shift towards training large language models (LLMs) using DeepSeek-R1-Zero-style reinforcement learning (RL) on verifiable rewards has led to impressive advancements in code and mathematical reasoning. However, this methodology is limited to tasks where rule-based answer verification is possible and does not naturally extend to real-world domains such as chemistry, healthcare, engineering, law, biology, business, and economics. Current practical workarounds use an additional LLM as a model-based verifier; however, this introduces issues such as reliance on a strong verifier LLM, susceptibility to reward hacking, and the practical burden of maintaining the verifier model in memory during training. To address this and extend DeepSeek-R1-Zero-style training to general reasoning domains, we propose a verifier-free method (**VeriFree**) that bypasses answer verification and instead directly maximizes the probability of generating the reference answer, derived in a principled way from the RL objective. We compare VeriFree with verifier-based methods and demonstrate that, in addition to its significant practical benefits and reduced compute requirements, VeriFree matches and even surpasses verifier-based methods on extensive evaluations across MMLU-Pro, GPQA, SuperGPQA, and math-related benchmarks.

The expansion of generative AI services highlights the growing need for adaptive mechanisms to select an appropriate generative model for responding to a user's prompts. Recent works have proposed offline and online learning formulations to identify the optimal generative model for an input prompt, based solely on maximizing prompt-based fidelity evaluation scores, e.g., CLIP-Score in text-to-image generation. However, such fidelity-based selection methods overlook the diversity of generated outputs, and hence, they can fail to address potential diversity shortcomings in the generated responses. In this paper, we introduce the *Diversity-Aware Kernelized Upper Confidence Bound (DAK-UCB)* method as a contextual bandit algorithm for the online selection of generative models with diversity considerations. The proposed DAK-UCB method incorporates both fidelity and diversity-related metrics into the selection process. We design this framework based on prompt-aware diversity score functions that decompose to a two-sample-based expectation over prompt-output pairs in the previous generation rounds. Specifically, we illustrate the application of our framework using joint kernel distance and kernel entropy measures. Our experimental results demonstrate the effectiveness of DAK-UCB in promoting diversity-aware model selection while maintaining fidelity in the generations for a sequence of prompts.

We study kernel regression with common rotation-invariant kernels on real datasets including CIFAR-5m, SVHN, and ImageNet. We give a theoretical framework that predicts learning curves (test risk vs. sample size) from only two measurements: the empirical data covariance matrix and an empirical polynomial decomposition of the target function $f_*$. The key new idea is an analytical approximation of a kernel’s eigenvalues and eigenfunctions with respect to an anisotropic data distribution. The eigenfunctions resemble Hermite polynomials of the data, so we call this approximation the \textit{Hermite eigenstructure ansatz} (HEA). We prove the HEA for Gaussian data, but we find that real image data is often ``Gaussian enough’’ for the HEA to hold well in practice, enabling us to predict learning curves by applying prior results relating kernel eigenstructure to test risk. Extending beyond kernel regression, we empirically find that MLPs in the feature-learning regime learn Hermite polynomials in the order predicted by the HEA. Our HEA framework is a proof of concept that an end-to-end theory of learning which maps dataset structure all the way to model performance is possible for nontrivial learning algorithms on real datasets.

As coding agents are increasingly deployed in large codebases, the need to automatically design challenging, codebase-level evaluation is central. We propose Gistify, a task where a coding LLM must create a single, minimal, self-contained file that can reproduce a specific functionality of a codebase. The coding LLM is given full access to a codebase along with a specific entrypoint (e.g., a python command), and the generated file must replicate the output of the same command ran under the full codebase, while containing only the essential components necessary to execute the provided command. Success on Gistify requires both structural understanding of the codebase, accurate modeling of its execution flow as well as the ability to produce potentially large code patches. Our findings show that current state-of-the-art models struggle to reliably solve Gistify tasks, especially ones with long executions traces.

In this paper, we present DiffuDETR, a novel approach that formulates object detection as a conditional object query generation task, conditioned on the image and a set of noisy reference points. We integrate DETR-based models with denoising diffusion training to generate object queries' reference points from a prior gaussian distribution. We propose two variants: DiffuDETR, built on top of the Deformable DETR decoder, and DiffuDINO, based on DINO’s decoder with contrastive denoising queries (CDNs). To improve inference efficiency, we further introduce a lightweight sampling scheme that requires only multiple forward passes through the decoder. Our method demonstrates consistent improvements across multiple backbones and datasets, including COCO2017, LVIS, and V3Det, surpassing the performance of their respective baselines, with notable gains in complex and crowded scenes. Using ResNet-50 backbone we observe a +1.0 in COCO-val reaching 51.9 mAP on DiffuDINO compared to 50.9 mAP of the DINO. We also observe similar improvements on LVIS and V3DET datasets with +2.4 and +2.2 respectively. Code will be released upon acceptance.