Showing 1–50 of 618 results for author: Liang, D

SAPO: Single-Rollout Autoregressive Policy Optimization for Agentic Reinforcement Learning

Authors: Dayang Liang, Lang Feng, Bo An, Yunlong Liu

Abstract: Agentic reinforcement learning (RL) has become a critical stage in the post-training of large language models. Existing critic-free, group-relative methods estimate policy advantages from multiple rollouts, avoiding the substantial memory overhead of conventional proximal policy optimization (PPO) and achieving strong performance on long-horizon interactive tasks. Despite their success, recent stu… ▽ More Agentic reinforcement learning (RL) has become a critical stage in the post-training of large language models. Existing critic-free, group-relative methods estimate policy advantages from multiple rollouts, avoiding the substantial memory overhead of conventional proximal policy optimization (PPO) and achieving strong performance on long-horizon interactive tasks. Despite their success, recent studies revealed three limitations: (1) Lack explicit value generalization and effective temporal credit assignment; (2) Suffer from potential advantage collapse in long-horizon complex tasks; (3) Require a costly trade-off between sampling budget and policy performance. In this work, we propose Single-rollout Autoregressive Policy Optimization (SAPO), a low-memory and compute-efficient framework in which the policy and value functions share a single autoregressive backbone. SAPO exploits the autoregressive structure of LLMs to produce policy and value predictions at distinct causal boundaries with shared parameters, while independently optimizing the PPO objectives and auxiliary on-policy SARSA objectives. To robustly estimate the contribution of each turn, we further introduce a trajectory-level generalized advantage estimator that combines lambda-returns with batch normalization. Experiments across ALFWorld and WebShop with Qwen2.5-1.5B/7B show that SAPO trains stably and outperforms PPO and GRPO by mean +15.1 and +12.1 percentage points, respectively, while eliminating the memory cost of a separate critic model and reducing per-iteration runtime by 33.2% over PPO. △ Less

Submitted 20 August, 2026; originally announced August 2026.

UniVerse: Benchmarking and Enhancing LALMs on Culturally Inclusive Low-Resource Music Understanding

Authors: Ziya Zhou, Shangda Wu, Shenyang Xu, Yutong Zheng, Dafang Liang, Suin Chung, Danbinaerin Han, Junyan Jiang, Yongyi Zang, Ruibin Yuan, Rongxiu Zhong, Shilei Zhang, Junlan Feng, Jinglei Liu, Haotian Zhou, Zijin Li, Dasaem Jeong, Wei Xue, Yike Guo

Abstract: Recent advances in large audio-language models (LALMs) have significantly improved performance in tasks such as music captioning, genre classification, and sound event detection. However, limited attention has been paid to improving their adaptability across diverse musical traditions, particularly folk music rooted in distinct cultural contexts. Folk-music traditions are typically resource-scarce… ▽ More Recent advances in large audio-language models (LALMs) have significantly improved performance in tasks such as music captioning, genre classification, and sound event detection. However, limited attention has been paid to improving their adaptability across diverse musical traditions, particularly folk music rooted in distinct cultural contexts. Folk-music traditions are typically resource-scarce, unevenly represented across regions, and poorly documented. Even when such samples appear in large-scale pre-training, LALMs often fail to capture their structural and stylistic characteristics, partly due to the absence of dedicated evaluation protocols and training solutions. To address these limitations, we introduce UniVerse, a reproducible solution for low-resource music understanding. Specifically, we propose UniVerseBench, a benchmark of 5,042 Q&A pairs across more than 38 cultural and linguistic entities, constructed via an expert-guided yet highly automated pipeline. In parallel, we construct a fully automated, model-generated multi-turn dialogue training dataset UniVerseSet. By training LALMs on UniVerseSet, we systematically adapt and investigate representative multimodal imbalance learning strategies across both dense and Mixture-of-Experts (MoE) architectures. Experimental results indicate that fully automated data curation combined with imbalance-aware training yields non-trivial improvements, but models still struggle to capture fine-grained acoustic features, indicating a gap between surface-level alignment and deep musical comprehension. △ Less

Submitted 18 August, 2026; originally announced August 2026.

Comments: 21 pages, 7 figures, 8 tables

PlanPO: Group Planning-Aware Policy Optimization for Multi-Turn Agentic LLMs

Authors: Dayang Liang, Liyuan He, Xuan Feng, Shuxin Li, Bo An, Yunlong Liu

Abstract: Group-relative policy optimization has emerged as a key paradigm for training agentic large language models (LLMs) on multi-turn interactive tasks. However, most existing variants fail to distinguish advantages among successful trajectories even when these trajectories differ substantially in their interaction efficiency. For instance, circuitous successes are often assigned the identical outcome… ▽ More Group-relative policy optimization has emerged as a key paradigm for training agentic large language models (LLMs) on multi-turn interactive tasks. However, most existing variants fail to distinguish advantages among successful trajectories even when these trajectories differ substantially in their interaction efficiency. For instance, circuitous successes are often assigned the identical outcome reward, causing advantage collapse and severe performance bottlenecks. To this end, we propose Group Planning-aware Policy Optimization (PlanPO), a simple yet effective RL method for learning generalizable planning abilities beyond task-specific high-quality behavior patterns. Specifically, PlanPO introduces coarse-to-fine advantage signals, which capture the relative differences in trajectory-level lengths and turn-level response lengths conditioned on successful trajectories sampled for the same task. Within the group-relative optimization structure, this enables agents to actively learn generalizable and deliberate behaviors spanning interaction planning and textual generation from high-quality rollouts, without degenerating into vanilla length minimization. Experimentally, PlanPO improves over GRPO by 27.2\% on average across the challenging multi-turn benchmarks ALFWorld, WebShop, and SciWorld, outperforming recent powerful baselines while incurring negligible additional training cost. △ Less

Submitted 17 August, 2026; originally announced August 2026.

PERO: Efficient Robust Post-Training Foundation Models for Encrypted Traffic Classification

Authors: Wumei Du, Jiarong Wen, Kaiyu Zhang, Zi Yang, Yiqin Lv, Longfei Zhang, Dong Liang, Zheng Xie

Abstract: Encrypted traffic classification is vital for network security, yet real-world deployments are inherently sensitive to rare but high-loss errors such as misclassification of malicious traffic. The encrypted traffic foundation model, as a promising general-purpose technique, can achieve impressive overall performance. However, employing standard objectives such as empirical risk minimization often… ▽ More Encrypted traffic classification is vital for network security, yet real-world deployments are inherently sensitive to rare but high-loss errors such as misclassification of malicious traffic. The encrypted traffic foundation model, as a promising general-purpose technique, can achieve impressive overall performance. However, employing standard objectives such as empirical risk minimization often overlooks high-risk tail events, and commonly used performance metrics hardly reflect robustness limitations in risk-sensitive scenarios. Directly applying robust optimization objectives, such as conditional value-at-risk, to post-training is computationally prohibitive for large models, as identifying high-loss samples exhausts substantial computation. To this end, we propose Pre-Evaluation Robust Optimization (PERO), an efficient robust post-training framework for encrypted traffic foundation models. PERO employs a lightweight proxy to estimate sample-wise risk and selects a subset of high-risk samples to update the foundation model, decoupling risk estimation from expensive large-model optimization. Extensive experiments on typical encrypted traffic datasets show that PERO achieves competitive or superior robustness and average performance compared to outstanding robust post-training methods, while significantly reducing computational and memory costs. △ Less

Submitted 15 August, 2026; originally announced August 2026.

Comments: 16 pages, 6 figures, 6 tables, conference

MSC Class: 68T07 (Primary) 68T20; 68T01; 68T10 (Secondary) ACM Class: I.5.1

WeSCE: A Benchmark for Measuring Security Drift in LLM-Driven Code Editing

Authors: Zhiyu Zhang, Tingyue Wen, Senke Sun, Dengxiang Liang, Enhao Huang

Abstract: In this work, we introduce WeSCE, a benchmark for quantifying security drift in code editing under weak-security constraints, where tasks specify only functional objectives without explicit security requirements. WeSCE consists of 400 executable programs derived from real-world code, covering feature addition, feature removal, bug fixing, and refactoring. To quantify security drift, we propose a c… ▽ More In this work, we introduce WeSCE, a benchmark for quantifying security drift in code editing under weak-security constraints, where tasks specify only functional objectives without explicit security requirements. WeSCE consists of 400 executable programs derived from real-world code, covering feature addition, feature removal, bug fixing, and refactoring. To quantify security drift, we propose a continuous risk representation that aggregates heterogeneous vulnerability signals through a unified formulation, and define drift measures capturing changes in overall risk, worst-case severity, and vulnerability distribution under code transformations, providing a multi-scale view of security spanning average-case behavior to worst-case emphasis. △ Less

Submitted 15 August, 2026; originally announced August 2026.

Harmful Content Is Not Enough: Continuation Framing Moderates In-Context Emergent Misalignment

Authors: Peiyang Liu, Xi Wang, Ziqiang Cui, Di Liang, Wei Ye

Abstract: In-context learning (ICL) can induce emergent misalignment (EM), where narrow misaligned examples alter answers to unrelated questions. Existing prompts, however, conflate harmful-text exposure with an invitation to continue assistant behavior. We hold harmful answers fixed while varying their delivery as demonstrations, evidence, assistant history, or tool output. Across ten independently sampled… ▽ More In-context learning (ICL) can induce emergent misalignment (EM), where narrow misaligned examples alter answers to unrelated questions. Existing prompts, however, conflate harmful-text exposure with an invitation to continue assistant behavior. We hold harmful answers fixed while varying their delivery as demonstrations, evidence, assistant history, or tool output. Across ten independently sampled contexts, demonstration framing raises broad EM by $30$--$32$ percentage points on a susceptible Gemini model; the gap survives domain exclusion, semantic clustering, unseen questions, and four prompt templates. Format and length-matched controls show that harmful content is necessary but insufficient. A role times continuation factorial further reveals model-dependent provenance effects: Gemini follows both assistant and tool histories, whereas Grok largely resists tool-framed continuation. Several other frontier and open-weight models show no gap. Blinded human audits confirm every main contrast and show that the model judge underestimates active-condition failures. Thus continuation framing is a strong, model-dependent moderator of ICL-EM, not a universal consequence of harmful context. △ Less

Submitted 8 August, 2026; originally announced August 2026.

SimWAM: A Simple World Action Model for End-to-End Autonomous Driving

Authors: Zongchuang Zhao, Xin Zhou, Tianyang Xu, Zhengyang Sun, Kaixuan Zhou, Honglin Li, Dingkang Liang, Xiang Bai

Abstract: World-Action Models (WAMs) improve end-to-end autonomous driving by transferring video dynamics priors to action prediction, but existing methods incur costly test-time future imagination. We present SimWAM, a simple yet effective WAM that leverages future-video prediction as a training-time supervision signal. It co-trains a pretrained video expert and a lightweight action expert with joint flow… ▽ More World-Action Models (WAMs) improve end-to-end autonomous driving by transferring video dynamics priors to action prediction, but existing methods incur costly test-time future imagination. We present SimWAM, a simple yet effective WAM that leverages future-video prediction as a training-time supervision signal. It co-trains a pretrained video expert and a lightweight action expert with joint flow matching. An isolated attention mask keeps action prediction independent of future frames, allowing trajectory prediction without explicit future-frame generation at inference. Since the two experts share no parameters and interact only through a unified attention interface, the video backbone could be replaced and the action expert scaled independently without modifying the learning objective or inference pipeline. We further apply reinforcement learning to optimize a compositional driving reward beyond trajectory imitation. Our SimWAM achieves 91.5 PDMS on NAVSIM, surpasses state-of-the-art WAM-based planners with substantially lower latency, and transfers zero-shot to nuScenes. These results position SimWAM as a simple yet solid baseline that could readily benefit from advances in video generation for efficient autonomous driving. The code and model weights are available at https://github.com/H-EmbodVis/SimWAM/. △ Less

Submitted 17 August, 2026; v1 submitted 7 August, 2026; originally announced August 2026.

Comments: The code and model weights are available at https://github.com/H-EmbodVis/SimWAM/

Ensemble-Based Residual Tests of GW231123 across Waveform Models

Authors: Dicong Liang, Hai-Tian Wang, Junlin Qin, Zhan-Feng Mai, Tong Jiang, Yingjie Yang

Abstract: GW231123 is an exceptional gravitational wave event for which different waveform models yield significantly different inferred source parameters. Residual tests provide a direct way to assess whether each waveform model gives an adequate description of the observed signal. In this work, we extend the conventional residual-test methods by subtracting the 100 highest likelihood waveforms, rather tha… ▽ More GW231123 is an exceptional gravitational wave event for which different waveform models yield significantly different inferred source parameters. Residual tests provide a direct way to assess whether each waveform model gives an adequate description of the observed signal. In this work, we extend the conventional residual-test methods by subtracting the 100 highest likelihood waveforms, rather than only the maximum likelihood waveform for each model, thereby propagating waveform reconstruction uncertainty into the residual analysis. This ensemble-based approach turns the residual test from a single waveform diagnostic into a robustness test over the local high likelihood waveform manifold. We further perform injection tests to quantify the detectability of cross-model waveform discrepancies in realistic detector noise. The large-scale implementation of these analyses is made possible by the high speed and low computational cost of our residual testing framework, which is based on three goodness-of-fit tests: the Kolmogorov-Smirnov test, the Anderson-Darling test, and Pearson's chi-squared test. △ Less

Submitted 6 August, 2026; originally announced August 2026.

Comments: 22 pages, 10 figures. Comments are welcome

K-space Gaussian Representation for Parallel MRI

Authors: Yu Guan, Mingyu Hu, Jiale Hu, Zhuoxu Cui, Dong Liang, Qiegen Liu

Abstract: Accelerated magnetic resonance imaging (MRI) aims to recover the k-space signal from acquired measurements, where accurate estimation of missing samples is essential for high-fidelity reconstruction. Existing k-space reconstruction methods estimate missing samples through interpolation operators or structure priors defined on discrete sampling grids. Although these formulations effectively exploit… ▽ More Accelerated magnetic resonance imaging (MRI) aims to recover the k-space signal from acquired measurements, where accurate estimation of missing samples is essential for high-fidelity reconstruction. Existing k-space reconstruction methods estimate missing samples through interpolation operators or structure priors defined on discrete sampling grids. Although these formulations effectively exploit local interpolation relationships and global k-space redundancy, they reconstruct only discrete frequency coefficients and therefore do not explicitly model the underlying continuous signal. To overcome this limitation, we propose K-space Gaussian Representation (KGR), the first explicit continuous representation formulated directly in the native k-space domain. Rather than estimating unknown samples on discrete grids, KGR parameterizes the continuous signal using Gabor-Gaussian primitives with shared spatial geometry, yielding a compact representation that naturally preserves inter-coil correlations. Because unconstrained continuous fitting does not necessarily satisfy the intrinsic structural properties of multi-coil signal, the estimated representation is projected onto a low-rank manifold to enforce the algebraic constraints arising from smoothly varying phase and coil redundancy. A frequency-adaptive fitting strategy accommodates the heterogeneous characteristics of different k-space regions. Comprehensive validation across multiple datasets and sampling schemes shows consistent improvements over representative reconstruction baselines in both quantitative metrics and visual quality. These results suggest that explicit continuous parameterization of native k-space provides a principled framework for integrating continuous signal modeling with structured low-rank reconstruction. △ Less

Submitted 4 August, 2026; v1 submitted 29 July, 2026; originally announced August 2026.

ROAD: Reciprocal-Objective Alignment of Discriminative Semantics for 3D Shape Generation

Authors: Xiao Luo, Mingyang Du, Xin Zhou, Tianrui Feng, Xiwu Chen, Xiaofan Li, Jiangning Zhang, Dingkang Liang

Abstract: High-fidelity 3D generation predominantly relies on scaling model capacity and data, which incurs prohibitive computational costs. This paradigm typically requires learning geometry from scratch and overlooks the rich semantic and structural priors already encapsulated in discriminative 3D foundation models. We contend that leveraging the profound understanding of the 3D world possessed by these d… ▽ More High-fidelity 3D generation predominantly relies on scaling model capacity and data, which incurs prohibitive computational costs. This paradigm typically requires learning geometry from scratch and overlooks the rich semantic and structural priors already encapsulated in discriminative 3D foundation models. We contend that leveraging the profound understanding of the 3D world possessed by these discriminative models can significantly reduce generative cost. To this end, we propose ROAD, a framework that reduces the training cost of 3D generation by transferring these rich discriminative priors into diffusion transformers. To address the inherent semantic-structural heterogeneity between generative and discriminative latents, we introduce a reciprocal-objective alignment strategy. This method synergizes Holistic Semantic Condensing to enforce global semantic coherence and Structural Optimal Alignment, which is formulated as a bipartite matching problem to rigorously align microscopic geometric details between disparate latent spaces. The 3D foundation model is only used for training-time supervision of alignment and is not used at inference, incurring no additional inference cost. Compared with the industrial baseline Step1X-3D, the proposed ROAD achieves highly competitive generation performance with only 1.5% of the training data and significantly reduces training costs, effectively reducing the computational overhead of high-fidelity 3D generation. Code is available at https://github.com/H-EmbodVis/ROAD. △ Less

Submitted 30 July, 2026; originally announced July 2026.

TurboVLA: Real-Time Vision-Language-Action Model at 32 Hz on an RTX 4090 with <1 GB VRAM

Authors: Hengyi Xie, Chenfei Yao, Xianjin Wu, Yingying Zhu, Dingkang Liang, Xiang Bai, Han Ding

Abstract: Vision-language-action (VLA) models commonly adopt an LLM-centric $V \to L \to A$ pathway, where visual observations are projected into the representation space of a large language model before being decoded into robot actions. Although effective, this design incurs substantial computation and memory overhead at every policy invocation. In this work, we introduce TurboVLA, a new VLA paradigm that… ▽ More Vision-language-action (VLA) models commonly adopt an LLM-centric $V \to L \to A$ pathway, where visual observations are projected into the representation space of a large language model before being decoded into robot actions. Although effective, this design incurs substantial computation and memory overhead at every policy invocation. In this work, we introduce TurboVLA, a new VLA paradigm that reformulates the conventional $V \to L \to A$ pathway as a direct $V + L \to A$ mapping. Instead of using a large language model as the central interface between perception and action, TurboVLA independently encodes visual observations and language instructions, directly exchanges information between them through lightweight bidirectional vision-language interaction, and predicts continuous action chunks with a compact decoder. This simple design constructs task-conditioned representations directly from visual and linguistic features, significantly reducing the computational and memory costs of VLA inference. On LIBERO, TurboVLA achieves 97.7% average success with only 0.2B parameters, 31.2 ms inference latency, and 0.9 GB inference VRAM on a consumer-grade RTX 4090, matching or outperforming substantially larger VLA policies. These results establish TurboVLA as a simple and effective alternative to the prevailing LLM-centric VLA paradigm, offering a new perspective on how vision, language, and action can be connected for efficient robotic manipulation. Code is available at https://github.com/H-EmbodVis/TurboVLA. △ Less

Submitted 16 August, 2026; v1 submitted 29 July, 2026; originally announced July 2026.

Comments: Code is available at https://github.com/H-EmbodVis/TurboVLA

WorldCupArena: Fine-Grained Evaluation of Language Models and Deep-Research Agents on Football Forecasting

Authors: Zhaokai Wang, Tianlin Gui, Jiayuan Rao, Shangzhe Di, Yihong Tang, Dingli Liang

Abstract: Predicting a football match before kickoff requires more than knowing past results: a model must use changing information and make a clear prediction before the answer is available. We present WorldCupArena, a dynamic benchmark for language models and deep-research agents. The 2026 FIFA World Cup is its first evaluation, and the same process can be reused for future leagues and cups. Before each m… ▽ More Predicting a football match before kickoff requires more than knowing past results: a model must use changing information and make a clear prediction before the answer is available. We present WorldCupArena, a dynamic benchmark for language models and deep-research agents. The 2026 FIFA World Cup is its first evaluation, and the same process can be reused for future leagues and cups. Before each match, a model either receives a common evidence package or searches for information itself. It predicts the result and score, likely players and events, match statistics, and the outcome of the competition. After the match, these predictions are compared with the recorded result. We report result accuracy, exact-score accuracy, and a scoreline score that gives some credit when a predicted score is close but not exact, together with scores for the other prediction tasks. Across 104 matches and 13 systems, models with similar result accuracy differ more clearly on detailed predictions. Compared with betting-market and human-fan baselines, the best system shows only small gains in result and exact-score accuracy, but a clearer gain in Scoreline. New schedules can be added as they begin, allowing the benchmark to evaluate future models without using outcomes that are already known. Code, prompts, predictions, and evaluation scripts are open sourced at https://github.com/wzk1015/WorldCupArena. △ Less

Submitted 20 July, 2026; originally announced July 2026.

Nexus: Native Mesh Generation with Diffusion

Authors: Hanxiao Wang, Ying-Tian Liu, Yuan-Chen Guo, Qi-Yuan Feng, Zi-Xin Zou, Ding Liang, Biao Zhang, Yan-Pei Cao

Abstract: Generating high-quality triangle meshes is essential for film, gaming, and interactive 3D applications. Mainstream methods rely on mesh serialization and autoregressive processes, which stuggles in effective inference and is sensitive to error accumulation. In this paper, we present Nexus, a diffusion method that achieves holistic mesh generation via decoupled vertex and topology generation. First… ▽ More Generating high-quality triangle meshes is essential for film, gaming, and interactive 3D applications. Mainstream methods rely on mesh serialization and autoregressive processes, which stuggles in effective inference and is sensitive to error accumulation. In this paper, we present Nexus, a diffusion method that achieves holistic mesh generation via decoupled vertex and topology generation. First, we view mesh vertices as sparse voxels organized as an octree and adopt a diffusion model to generate the vertices in a coarse-to-fine manner. Second, for topology modeling, we propose Spacetime Interval, as an extension of Spacetime Distance to encode arbitrary edge and face topology into continuous per-vertex embeddings. It allows for a global and efficient recovery of complex topology. We then employ a diffusion model to generate the continuous embeddings on the generated vertices. Extensive experiments on the Objaverse and Toys4K datasets and in-the-wild images demonstrate that our method outperforms state-of-the-art autoregressive and two-stage baselines, effectively circumventing the inherent limitations of sequential mesh modeling. A blind user study from 3D practitioners confirms strong perceptual preference for our results. △ Less

Submitted 15 July, 2026; originally announced July 2026.

SeamGen: Artist-Aligned UV Seam Generation via Graph Flow Matching

Authors: Hao Xu, Yuqing Zhang, Yiqian Wu, Xueqi Ma, Ding Liang, Yan-Pei Cao, Ying-Tian Liu, Xiaogang Jin

Abstract: UV seam placement is a critical yet labor-intensive step in 3D content creation, requiring artists to balance chart shape, seam concealment, and alignment with semantic and geometric features. Existing automatic methods are primarily based on per-object optimization, relying on handcrafted objectives to avoid distortion or on proxies from pretrained models to inject semantic information. However,… ▽ More UV seam placement is a critical yet labor-intensive step in 3D content creation, requiring artists to balance chart shape, seam concealment, and alignment with semantic and geometric features. Existing automatic methods are primarily based on per-object optimization, relying on handcrafted objectives to avoid distortion or on proxies from pretrained models to inject semantic information. However, these strategies are not always well aligned with seams used in industrial production pipelines, often resulting in layouts that deviate from artist-preferred seam patterns and practical production requirements. To address these limitations, we propose SeamGen, a generative model for UV seam generation that aligns with artist preferences and production requirements. Instead of depending on manually designed objectives and constraints, SeamGen learns the distribution of per-edge seam labels from a large corpus of existing seam layouts using a flow-matching generative model. A key challenge is that typical Transformer architectures used in flow matching models are designed for sequential representations, such as point clouds, and cannot naturally account for mesh topology. To enable mesh-native learning, we design a Mesh Transformer backbone that interleaves local graph attention over mesh edges with global self-attention across vertices, capturing both fine-grained geometric cues and long-range topological coherence. To further improve inference-time controllability and quality, we exploit the training-free inpainting capability of flow models for both localized seam refinement and constraint-guided seam generation. Extensive experiments show that by learning priors from professional seam layout data, SeamGen produces UV layouts that better align with artist-authored preferences and achieve superior perceptual quality compared with distortion-based and semantic-proxy baselines. △ Less

Submitted 14 July, 2026; originally announced July 2026.

Wat3R: Underwater 3D Geometry Learning without Annotations

Authors: Jiangwei Ren, Xingyu Jiang, Zijie Song, Wei Xu, Hongkai Lin, Dingkang Liang, Xiang Bai

Abstract: Estimating 3D geometry in underwater environments presents unique challenges due to light attenuation, scattering, and the absence of large-scale, high-quality 3D annotations. Pioneering methods rely on massive dense annotations that are impractical in underwater settings. In this paper, we propose Wat3R, a cross-domain semi-supervised learning framework designed to adapt feed-forward 3D reconstru… ▽ More Estimating 3D geometry in underwater environments presents unique challenges due to light attenuation, scattering, and the absence of large-scale, high-quality 3D annotations. Pioneering methods rely on massive dense annotations that are impractical in underwater settings. In this paper, we propose Wat3R, a cross-domain semi-supervised learning framework designed to adapt feed-forward 3D reconstruction models from air to underwater scenes. Uniquely, our method eliminates the need for any annotated underwater data following a teacher-student architecture, that learns robust geometry representations merely on abundant unlabeled real underwater video footage. We also design a cross-view consistency loss that leverages geometric cues from other views to compensate for the information degradation in the current view caused by water attenuation and scattering. Furthermore, considering the lack of comprehensive evaluation benchmarks, we construct Water3D, a diverse dataset covering various water bodies and underwater scenarios, designed for geometric task evaluation. Experimental results demonstrate that Wat3R outperforms current state-of-the-art methods in underwater multi-view depth estimation and point cloud reconstruction. The dataset and code are available at https://github.com/LSXI7/Wat3R . △ Less

Submitted 9 July, 2026; originally announced July 2026.

Comments: Accepted to ECCV 2026. The dataset and code are available at https://github.com/LSXI7/Wat3R

Multi-wavelength Constraints on the Transient EP250905a

Authors: J. Quirola-Vasquez, P. G. Jonker, A. Levan, D. B. Malesani, F. E. Bauer, A. Martin-Carrillo, G. Corcoran, D. Mata Sanchez, R. A. J. Eyles-Ferris, F. Carotenuto, M. Ravasio, J. Sanchez-Sierras, J. Bright, J. A. Chacon, L. Cotter, F. J. Cowie, N. Sarin, M. A. P. Torres, J. N. D. van Dalen, A. P. C. van Hoof, V. D'Elia, P. Jakobsson, N. Habeeb, S. Kobayashi, A. Saccardi , et al. (8 additional authors not shown)

Abstract: Fast X-ray transients (FXTs) are a diverse class of high-energy suggested origins, ranging from stellar explosions to compact object mergers. The Einstein Probe (EP) satellite discovers approximately 100 FXTs per year. We aim to constrain the physical origin of EP250905a. We analyze X-ray, optical, near-infrared (NIR), and radio temporal and spectral properties of EP250905a. In addition, we assess… ▽ More Fast X-ray transients (FXTs) are a diverse class of high-energy suggested origins, ranging from stellar explosions to compact object mergers. The Einstein Probe (EP) satellite discovers approximately 100 FXTs per year. We aim to constrain the physical origin of EP250905a. We analyze X-ray, optical, near-infrared (NIR), and radio temporal and spectral properties of EP250905a. In addition, we assess the possible role of weak gravitational lensing in shaping its observed characteristics. EP250905a fades rapidly in X-rays, and we detect no NIR or radio emission, but we detect early optical emission that rapidly fades beyond the detection limits. Two nearby galaxies are identified for which we derive spectroscopic redshifts of $z=0.374$ (G1) and $z=2.714$ (G2). Our analysis favors G2 as the host of the FXT EP250905a. The angular separation of 2.56\arcsec\, between the FXT's optical counterpart and the center of the G1 galaxy suggests the emission of the FXT might be moderately magnified by lensing effects ($μ\approx3.9$) given the inferred Einstein radius of G1 ($θ_E\approx1.9$ arcsec). The data are best explained as an afterglow from a mildly off-axis structured jet at $z=2.714$, providing a consistent broadband interpretation that also allows for weakly lensed emission of EP250905a. △ Less

Submitted 8 July, 2026; originally announced July 2026.

Comments: Accepted for publication in A&A

Demonstration and Design of Uni-Directional and Ultra-Low Threshold Hybrid Quantum Dot III-V/Si Micro-Ring Laser

Authors: Xucheng Yang, Yingtao Hu, Antoine Descos, Yuan Yuan, Bassem Tossoun, Geza Kurczveil, Yatiraj Ramanujam, Jonathan Wierer, Raymond G. Beausoleil, Di Liang, Stanley Cheung

Abstract: Micro-ring lasers (MRLs) are attractive light sources for energy-efficient optical interconnects, but their intrinsic directional bistability leads to unpredictable clockwise/counter-clockwise emission. We demonstrate stable unidirectional emission in hybrid quantum-dot (QD) III-V/Si MRLs using passive reflective feedback integrated on the bus waveguide, leaving the ring cavity unperturbed. Three… ▽ More Micro-ring lasers (MRLs) are attractive light sources for energy-efficient optical interconnects, but their intrinsic directional bistability leads to unpredictable clockwise/counter-clockwise emission. We demonstrate stable unidirectional emission in hybrid quantum-dot (QD) III-V/Si MRLs using passive reflective feedback integrated on the bus waveguide, leaving the ring cavity unperturbed. Three reflector architectures - Y-splitter loop mirrors, adiabatic Y-splitter loop mirrors, and distributed Bragg reflectors (DBRs) - are benchmarked against a reflector-free bidirectional baseline through combined experiment and coupled-mode-theory rate-equation modeling. All designs preserve ultra-low thresholds of 0.79-1.12 mA (112-158 A/cm^2, roughly an order of magnitude below prior quantum-well unidirectional ring lasers) while enhancing single-facet output power and wall-plug efficiency, with directional isolation up to 27.65 dB for the DBR. The reflectors impose no penalty on the 4-5 GHz modulation bandwidth or its thermal robustness, establishing passive external feedback as a practical route to unidirectional QD MRLs for DWDM-scale optical interconnects. △ Less

Submitted 2 July, 2026; originally announced July 2026.

On the Power Set of Quasinilpotent Operators in Banach Spaces

Authors: C. L. Hu, Y. Q. Ji, D. H. Liang

Abstract: For a quasinilpotent operator $T$ on a Banach space $X$, Douglas and Yang defined $k_{x}=\limsup\limits_{λ\rightarrow 0}\frac{\ln\|(λ-T)^{-1}x\|}{\ln\|(λ-T)^{-1}\|}$ for each non-zero vector $x$, and called $Λ(T)=\{k_x:x\neq 0\}$ the power set of $T$. In this paper, we prove that $Λ(T)$ always contains $1$ for every quasinilpotent operator $T$ on $X$. Moreover, we introduce the concept… ▽ More For a quasinilpotent operator $T$ on a Banach space $X$, Douglas and Yang defined $k_{x}=\limsup\limits_{λ\rightarrow 0}\frac{\ln\|(λ-T)^{-1}x\|}{\ln\|(λ-T)^{-1}\|}$ for each non-zero vector $x$, and called $Λ(T)=\{k_x:x\neq 0\}$ the power set of $T$. In this paper, we prove that $Λ(T)$ always contains $1$ for every quasinilpotent operator $T$ on $X$. Moreover, we introduce the concept of a Banach space $X$ having uniform multiplicity infinity and prove that some classical Banach spaces possess this property. As an application, we show that if $σ\subset [0,1]$ is right closed and contains $1$, then there exists a quasinilpotent operator $T$ on a class of Banach spaces with uniform multiplicity infinity such that $Λ(T)=σ$. △ Less

Submitted 1 July, 2026; originally announced July 2026.

MSC Class: Primary 47A10; Secondary 47B37

High-dimensional Embedding Prior for Noisy K-space Domain MRIReconstruction

Authors: Yu Guan, Tianjia Huang, Qinrong Cai, Qiuyun Fan, Dong Liang, Qiegen Liu

Abstract: Magnetic resonance imaging (MRI) reconstruction under realistic acquisition conditions can be fundamentally viewed as estimating the underlying k-space distribution from incomplete and noise-corrupted measurements. While diffusion models have recently shown strong potential as generative prior for inverse problems,existingapproachesstruggletohandlenoisyreconstruction settings, especially when oper… ▽ More Magnetic resonance imaging (MRI) reconstruction under realistic acquisition conditions can be fundamentally viewed as estimating the underlying k-space distribution from incomplete and noise-corrupted measurements. While diffusion models have recently shown strong potential as generative prior for inverse problems,existingapproachesstruggletohandlenoisyreconstruction settings, especially when operating directly in k-space domain. In this work, we propose a unified high-dimensional k-space reconstruction framework tailored for noisy inverse problems, whichenhancesdiffusion-based solversthroughrepresentation lifting.Ratherthanmodifyingthe underlying optimization procedures, the proposed framework augments the data representation space, enabling existing diffusion-based solvers to operate on enriched k-space embeddings with improved expressiveness. Extensive experiments on both in-house and public datasets across varying noise levels and undersampled factors demonstrate that the proposed frame work consistently improves reconstruction quality for multiple diffusion-based inverse solvers. Notably, the largest gains are observed in high-noise regimes, which is consistent with our theoretical analysis of error propagation under high-dimensional representation. These results suggest that high-dimensional representation provides a general and model-agnostic mechanism for improving diffusion-based MRI reconstruction in noisy settings, offering a new perspective on robust k-space generative modeling for practical inverse problems. The code will be available at https://github.com/yqx7150/HEP-MRIRec. △ Less

Submitted 1 July, 2026; originally announced July 2026.

Mixture of Debaters: Learn to Debate at Architectural Level in Multi-Agent Reasoning

Authors: Dayong Liang, Kaisong Gong, Yi Cai, Changmeng Zheng, Xiao-Yong Wei

Abstract: Existing multi-agent debate frameworks suffer from two critical limitations: they rely on static architectures where agent roles and coordination patterns are fixed at design time, and they require instantiating multiple model copies, incurring substantial computational overhead. We propose Mixture of Debaters (MoD), a unified framework that enables dynamic self-debate within a single model by lev… ▽ More Existing multi-agent debate frameworks suffer from two critical limitations: they rely on static architectures where agent roles and coordination patterns are fixed at design time, and they require instantiating multiple model copies, incurring substantial computational overhead. We propose Mixture of Debaters (MoD), a unified framework that enables dynamic self-debate within a single model by leveraging the Mixture-of-Experts paradigm. We address three key challenges in adapting MoE for dialectical reasoning: (1) dual-routing that decouples role allocation from process flow, dynamically determining when to debate versus when to synthesize; (2) momentum switching that smooths token-level routing with local context, reducing expert-switch jitter; and (3) unified self-debate that encapsulates diverse debating personas into lightweight expert modules, eliminating inter-agent communication while preserving behavioral diversity. Extensive experiments on multimodal benchmarks demonstrate that MoD outperforms both single-model baselines and conventional multi-agent systems, achieving superior accuracy with 3.7x lower latency and 87% reduction in token consumption.The source code can be accessed at https://github.com/YongLD/MoD. △ Less

Submitted 28 June, 2026; originally announced June 2026.

A pilot study examining transcranial photobiomodulation therapy intervention in college students with insomnia

Authors: Jiangshan He, Lianghua Zhang, Dan Liang, Xiaoyu Wang, Tianyi Luo, Haoda Wang, Ziqi Ren, Mingzhe Jiang, Lei Zheng, Qiyuan Cheng, Hui Xie, Xueli Chen

Abstract: College students commonly report insufficient sleep and poor sleep quality, with ~30% meeting insomnia criteria, posing significant threats to their physical growth, cognitive development, and overall well-being, as well as imposing a substantial economic burden on society [1]. The hyperarousal model of insomnia [2] emphasizes that hyperarousal across cognitive, emotional, and physiological domain… ▽ More College students commonly report insufficient sleep and poor sleep quality, with ~30% meeting insomnia criteria, posing significant threats to their physical growth, cognitive development, and overall well-being, as well as imposing a substantial economic burden on society [1]. The hyperarousal model of insomnia [2] emphasizes that hyperarousal across cognitive, emotional, and physiological domains mutually reinforces one another. Neuroimaging studies have further identified prefrontal hypoactivity as a key neural substrate underlying these dysfunctional cognitions and elevated arousal, reflecting a failure of top-down modulatory control over both limbic reactivity [3] and brainstem arousal nuclei [4]. Moreover, transcranial photobiomodulation (tPBM) therapy targeting the prefrontal cortex has demonstrated therapeutic efficacy across neuropsychiatric disorders with insomnia comorbidities [5,6], providing preliminary support for its application in insomnia. However, the neuro mechanisms underlying tPBM's therapeutic effects on insomnia remain to be elucidated. △ Less

Submitted 23 June, 2026; originally announced June 2026.

Comments: 1 figure

SciRisk-Bench: A Risk-Dimension-Aware Benchmark for AI4Science Safety

Authors: Linghao Feng, Yinqian Sun, Dongqi Liang, Sicheng Shen, Chenfei Yan, Yuxuan Peng, Yilin Zhao, Haibo Tong, Kai Li, FeiFei Zhao, Yi Zeng

Abstract: Large language models (LLMs) are increasingly embedded in AI for Science (AI4Science) workflows, from scientific question answering and literature analysis to laboratory planning and autonomous discovery. This progress creates an urgent need for safety benchmarks that evaluate not only scientific competence, but also whether models recognize and avoid risks in high-stakes scientific contexts. Exis… ▽ More Large language models (LLMs) are increasingly embedded in AI for Science (AI4Science) workflows, from scientific question answering and literature analysis to laboratory planning and autonomous discovery. This progress creates an urgent need for safety benchmarks that evaluate not only scientific competence, but also whether models recognize and avoid risks in high-stakes scientific contexts. Existing AI4Science safety datasets cover several disciplines and task formats, leaving the underlying risk dimensions underspecified. We introduce \textbf{SciRisk-Bench}, a benchmark designed to evaluate AI4Science safety from two complementary perspectives: explicit risk dimensions and scientific disciplines. SciRisk-Bench covers 7 disciplines, 31 subdisciplines and 10 risk dimensions. In the experimental section, we evaluate both mainstream LLMs and science-oriented LLMs across risk dimensions, disciplines, and sub-disciplines, enabling fine-grained diagnosis of where scientific models remain unsafe. △ Less

Submitted 24 June, 2026; v1 submitted 17 June, 2026; originally announced June 2026.

EgoCS-400K: An Egocentric Gameplay Dataset for World Models

Authors: Rongjin Guo, Dong Liang, Yuhao Liu, Fang Liu, Tianyu Huang, Gerhard P. Hancke, Rynson W. H. Lau

Abstract: The shift from video generation to interactive world modeling places new demands on data: beyond captioned videos, world models require temporally aligned video-action-language trajectories grounded in the actions, camera motion, states, and events that drive future scene changes. However, such data is difficult to obtain at scale. Web video datasets offer broad visual coverage but lack executable… ▽ More The shift from video generation to interactive world modeling places new demands on data: beyond captioned videos, world models require temporally aligned video-action-language trajectories grounded in the actions, camera motion, states, and events that drive future scene changes. However, such data is difficult to obtain at scale. Web video datasets offer broad visual coverage but lack executable actions and reliable states; robotic datasets provide action and state supervision but are costly and limited in scene diversity; and existing simulators often lack large-scale human-driven interaction trajectories. In this paper, we introduce EgoCS-400K, a large-scale replay-grounded egocentric Counter-Strike dataset for world models, built from public professional CS and CS2 match demos that preserve human gameplay trajectories and enable parsing, replaying, rendering, and temporal alignment. We extract player states, view directions, movements, keyboard/button inputs, view-angle changes, weapon usage, game events, and round-level context, and render clean first-person videos from the same trajectories. EgoCS-400K contains over 400,000 first-person videos and 10,000 hours of gameplay from more than 1,000 matches and 40,000 rounds, covering 13 maps and 10 player viewpoints per round. It supports a range of interactive visual modeling tasks, including action-conditioned future prediction, state- and event-aware scene rollout, replay-grounded captioning, and agent egocentric action understanding. By connecting visual observations with human actions, camera motion, game states, and events at scale, EgoCS-400K serves as a practical bridge between passive web videos, controllable game simulation, and costly real-world embodied data. △ Less

Submitted 16 June, 2026; originally announced June 2026.

Experimental Design Space Exploration of Ultra-Low Threshold Hybrid III-V/Si Quantum Dot Microring Lasers

Authors: Xucheng Yang, Preston Luong, Yatiraj Ramanujam, Antoine Descos, Yingtao Hu, Yuan Yuan, Bassem Tossoun, Geza Kurczveil, Eunso Shin, Jonathan Wierer, Ray Beausoleil, Di Liang, Stanley Cheung

Abstract: In this work, we report on the design strategies and experimental validation of ultra-low threshold ($< 0.8\,\mathrm{mA}$) hybrid III--V/Si quantum dot (InAs/GaAs) micro-ring lasers with optical output powers $> 2\,\mathrm{mW}$ for $1.3\,μ\mathrm{m}$ emission. The multi-dimensional design exploration allows for the demonstration of record wall-plug efficiencies ($\sim 10\%$) and threshold current… ▽ More In this work, we report on the design strategies and experimental validation of ultra-low threshold ($< 0.8\,\mathrm{mA}$) hybrid III--V/Si quantum dot (InAs/GaAs) micro-ring lasers with optical output powers $> 2\,\mathrm{mW}$ for $1.3\,μ\mathrm{m}$ emission. The multi-dimensional design exploration allows for the demonstration of record wall-plug efficiencies ($\sim 10\%$) and threshold current densities ($109\,\mathrm{A/cm^2}$) for these compact sources on silicon. We also demonstrate the thermal performance of several designs with record characteristic temperature values of $T_0 = 212\,\mathrm{K}$, indicating minimal temperature dependence of the threshold current. In addition, the high differential gain allows for the demonstration of 3-dB bandwidths up to $5\,\mathrm{GHz}$. △ Less

Submitted 11 June, 2026; originally announced June 2026.

Atomic Intent Reasoning: Bringing LLM Semantics to Industrial Cross-Domain Recommendations

Authors: Zhuohang Jiang, Yuxin Chen, Shijie Wang, Haohao Qu, Zhou Jindong, Wenqi Fan, Li Qing, Dongxu Liang, Jun Wang

Abstract: Cross-domain recommendation is a core problem in content-to-e-commerce platforms. Its objective is to leverage user interactions with content to infer potential purchasing intent on the e-commerce side, thereby enhancing conversion rates and commercial value. However, in real industrial scenarios, cross-domain recommendation faces multiple challenges: significant semantic gaps exist between differ… ▽ More Cross-domain recommendation is a core problem in content-to-e-commerce platforms. Its objective is to leverage user interactions with content to infer potential purchasing intent on the e-commerce side, thereby enhancing conversion rates and commercial value. However, in real industrial scenarios, cross-domain recommendation faces multiple challenges: significant semantic gaps exist between different domains, and user cross-domain behavior sequences are often massive in scale and rich in noise. Although large language models (LLMs) possess powerful semantic understanding and reasoning capabilities, their millisecond-level inference latency makes direct application in online recommendation systems difficult. To address these issues, this paper introduces AIR (Atomic Intent Reasoning), an LLM-driven cross-domain recommendation framework designed for industrial-grade deployment. By migrating LLM inference to the offline phase and dynamically constructing user intent representations through efficient retrieval and composition during online operations, it achieves approximately 400* inference acceleration while maintaining semantic consistency. Experimental results across multiple public datasets demonstrate that our method achieves state-of-the-art performance in cross-domain recommendation tasks. Furthermore, large-scale online A/B testing conducted in Kuaishou E-commerce's real-world business scenarios shows that our approach delivers stable and significant improvements across multiple core business metrics, including a +3.446% increase in GMV, fully validating its effectiveness and practical value in industrial-scale recommendation systems. △ Less

Submitted 8 June, 2026; originally announced June 2026.

Journal ref: Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2 (KDD '26), August 09--13, 2026, Jeju Island, Republic of Korea

UXBench: Benchmarking User Experience in AI Assistants

Authors: Mengze Hong, Xia Zeng, Zeyang Lei, Sheng Wang, Chen Jason Zhang, Di Jiang, Taiming Fu, Jinfeng Huang, Mengqiao Liu, Qinghe Chang, Haosheng Zou, Qiongyi Zhou, Sijun He, Simonjmdeng, Haojing Huang, Zijian Li, Lucas Mu Li, Fubao Zhang, Mona Zhou, Wei Ma, Yuan Hua, Qi Zhu, Shuo Jiang, Chenxuan Ma, Yuanmeng Zhang , et al. (4 additional authors not shown)

Abstract: As AI assistants serve millions of users daily, evaluating user experience (UX) beyond general model capability has become increasingly important. We present UXBench, the first user-centric benchmark grounded in real user feedback signals for evaluating preference alignment and dialogue generation. The benchmark consists of three interconnected tasks, UX Judge, UX Eval, and UX Recovery, with 7,400… ▽ More As AI assistants serve millions of users daily, evaluating user experience (UX) beyond general model capability has become increasingly important. We present UXBench, the first user-centric benchmark grounded in real user feedback signals for evaluating preference alignment and dialogue generation. The benchmark consists of three interconnected tasks, UX Judge, UX Eval, and UX Recovery, with 7,400 test instances extracted from over 70K interaction logs of a mainstream Chinese AI assistant. The dataset closely reflects real user distributions, covering 8 scenarios, 83 domains, and diverse failure patterns that pose severe challenges. Extensive experiments on 26 frontier language models provide novel insights into how well models perceive user experience and how improvements in model capability contribute to better dialogue engagement. Through comprehensive analysis of model behavior and performance gaps, we show that user feedback prediction is a learnable capability, where a reward model trained from in-the-wild feedback signals can achieve well-calibrated accuracy. We further document the systematic biases of LLM-as-a-judge evaluation protocols and compare typical response strategies that directly affect user experience. UXBench establishes a new evaluation landscape and calls for greater attention to tailored UX optimization, contributing to a user-centric scaling law that shapes the success of AI assistants. △ Less

Submitted 14 July, 2026; v1 submitted 8 June, 2026; originally announced June 2026.

ForesightSafety-SAGE:A Fully Automated Scenario Generation and Safety Evaluation Framework for LLM Agents

Authors: Lu Jia, Haibo Tong, Feifei Zhao, Jindong Li, Dongqi Liang, Ping Wu, Qian Zhang, Yi Zeng

Abstract: Large language models (LLMs) are increasingly evolving from simple text-based interaction systems into LLM agents that can maintain memory, use tools, access external environments, and execute tasks. As their capabilities and autonomy expand, the safety risks they face also become more diverse. Existing evaluations often rely on manually written scenarios, static prompts, or final-output judgments… ▽ More Large language models (LLMs) are increasingly evolving from simple text-based interaction systems into LLM agents that can maintain memory, use tools, access external environments, and execute tasks. As their capabilities and autonomy expand, the safety risks they face also become more diverse. Existing evaluations often rely on manually written scenarios, static prompts, or final-output judgments, making it difficult to capture the diverse risks that agents may face during task execution. We introduce ForesightSafety-SAGE, a fully automated scenario generation and safety evaluation framework for LLM agents. Based on five risk dimensions,we instantiae abstract and diverse safety risks in real-world task execution into 1,072 measurable evaluation scenarios. Using the automated evaluation pipeline, 12 LLM agents are evaluated under two authority contexts. The results show that current agents still face substantial behavioral safety risks during task execution, with an average ASR of 47.1% and several models exceeding 70%. These findings demonstrate the importance of executable, process-level evaluation for understanding and improving LLM agent safety. △ Less

Submitted 7 August, 2026; v1 submitted 7 June, 2026; originally announced June 2026.

Asymptotically-flat Black holes in Bumblebee gravity: Exact solutions and Thermodynamics

Authors: Jinbo Yang, Zhan-Feng Mai, Dicong Liang, Lijing Shao

Abstract: We construct analytic solutions to the bumblebee gravity theory in static and spherically symmetric spacetimes, where the bumblebee vector field admits only a non-vanishing temporal component. In particular, we identify the parameter space that allows for asymptotically flat black hole solutions. We further investigate the thermodynamic properties of these black holes and obtained the analytic for… ▽ More We construct analytic solutions to the bumblebee gravity theory in static and spherically symmetric spacetimes, where the bumblebee vector field admits only a non-vanishing temporal component. In particular, we identify the parameter space that allows for asymptotically flat black hole solutions. We further investigate the thermodynamic properties of these black holes and obtained the analytic formulas for the $Y$ charge and $X$ potential, which were introduced in the prior work to ensure the Smarr relation and the first law of black hole thermodynamics. Using the new analytic results, we verify the numerical findings reported in early work and uncover multiple cases missed in the previous numerical analysis. These include: (i) an unbounded charge-mass ratio when the non-minimal coupling parameter $ξ$ is larger than $2κ$, (ii) the emergence of a traversable wormhole configuration for overcharged solutions with $ξ<0$, (iii) the non-monotonic turning behavior of the Hawking temperature as a function of the charge-mass ratio, and (iv) the presence of two divergent points in the constant-$Y$ heat capacity. △ Less

Submitted 4 June, 2026; originally announced June 2026.

Comments: 22 pages,14 figures

Multi-Granularity Reasoning for Natural Language Inference

Authors: Chunling Xi, Di Liang

Abstract: Natural Language Inference (NLI) is a fundamental task in natural language understanding that requires determining the logical relationship between a premise and a hypothesis. Despite the remarkable success of transformer-based pre-trained models, most existing approaches primarily rely on the final-layer token representations, which are often insufficient for capturing the complex and hierarchica… ▽ More Natural Language Inference (NLI) is a fundamental task in natural language understanding that requires determining the logical relationship between a premise and a hypothesis. Despite the remarkable success of transformer-based pre-trained models, most existing approaches primarily rely on the final-layer token representations, which are often insufficient for capturing the complex and hierarchical semantic interactions required for effective reasoning. In particular, fine-grained lexical cues, phrasal compositions, and higher-level contextual semantics are typically entangled or diluted in a single representation space. To address these limitations, we propose a novel \emph{Multi-Granularity Reasoning Network} (MGRN) that explicitly leverages hierarchical semantic features within an interactive reasoning space. The proposed framework mimics the human cognitive process of language understanding, which naturally progresses from shallow lexical matching to deeper semantic abstraction and logical reasoning. By integrating semantic information across multiple granularities in a progressive and structured manner, MGRN is able to uncover intricate semantic relationships underlying natural language expressions. Extensive experiments on multiple public benchmarks demonstrate that MGRN consistently outperforms strong baseline models, validating the effectiveness and robustness of the proposed approach. △ Less

Submitted 22 June, 2026; v1 submitted 18 April, 2026; originally announced June 2026.

Efficient Hyperparameter Optimization for LLM Reinforcement Learning

Authors: Minping Chen, Bowen Xiao, Du Liang, Chuxuan Zeng, Zeyi Wen

Abstract: Reinforcement learning (RL) for large language models (LLMs) is highly sensitive to hyperparameter configurations, making hyperparameter optimization (HPO) essential yet computationally expensive. Existing multi-fidelity HPO methods remain inefficient for LLM RL due to the massive model scale and resource-intensive training cycles. In this paper, we propose Joint Fidelity Hyperparameter Optimizati… ▽ More Reinforcement learning (RL) for large language models (LLMs) is highly sensitive to hyperparameter configurations, making hyperparameter optimization (HPO) essential yet computationally expensive. Existing multi-fidelity HPO methods remain inefficient for LLM RL due to the massive model scale and resource-intensive training cycles. In this paper, we propose Joint Fidelity Hyperparameter Optimization (JF-HPO), which simultaneously adapts both model size and training budget as fidelity. JF-HPO is empowered by: (i) it leverages a small proxy model of the target LLM for efficient training and evaluation in each HPO trial; (ii) it integrates carefully designed early-stopping strategies based on training dynamics; (iii) it introduces an efficient checkpointing mechanism to eliminate redundant computations. Compared with existing HPO methods, JF-HPO significantly improves the computational efficiency of each trial (up to 14.9 times), while achieving better or competitive predictive accuracy under the same time budget. Notably, compared with utilizing hyperparameter configurations from the VeRL Recipe, JF-HPO delivers performance improvements ranging from 5.8% to 111.6%. △ Less

Submitted 1 June, 2026; originally announced June 2026.

Comments: 12 pages, 6 figures, accepted at ACL 2026

TRACE: Discovering Task-Specific Parameter via Adaptation-Aware Probing for Continual Fine-Tuning

Authors: Xiaosong Han, Ke Chen, Xindi Dai, Di Liang, Minlong Peng, Wei Pang, Fausto Giunchiglia, Xiaoyue Feng, Yonghao Liu, Renchu Guan

Abstract: In real-world deployment, LLMs are often adapted continually across tasks to keep LLMs up-to-date in production, where new fine-tuning should preserve previously learned skills. However, indiscriminately mixing tasks can dilute task specialization, while sequential fine-tuning (full-parameter or low rank adaptation) often causes catastrophic forgetting due to destructive overwriting. Replay-based… ▽ More In real-world deployment, LLMs are often adapted continually across tasks to keep LLMs up-to-date in production, where new fine-tuning should preserve previously learned skills. However, indiscriminately mixing tasks can dilute task specialization, while sequential fine-tuning (full-parameter or low rank adaptation) often causes catastrophic forgetting due to destructive overwriting. Replay-based continual tuning and maintaining separate task-specific adapters can mitigate forgetting, but introduce additional compute, storage, and management overhead. Recognizing the redundancy of LLM parameters for any single task, we reframe continual task adaptation as task-specific parameter discovery via adaptation-aware probing: a short warm-start probe exposes a task's adaptation trace, enabling us to identify and isolate the small subset of parameters essential for each task to mitigate catastrophic forgetting. Building on this view, we introduce TRACE, a novel approach for discovering Task-specific paRameters via Adaptation-aware probing for Continual finE-tuning. We perform a short warm-start fine-tune to derive task-specific core parameters by comparing the warm-started and pre-trained models. Core parameters are identified via two strategies: importance scoring (L$_2$ norm and Fisher Information) and specificity analysis (cosine similarity of parameter updates). In continual fine-tuning settings, only the active task's core parameters are updated while others remain frozen, preserving prior knowledge. We conduct extensive experiments across multiple standard benchmarks to demonstrate the superior performance of our proposed method. Additionally, we validate the generalization of our method through a cross-model and scale transferability study, demonstrating a "small-to-large" paradigm that guides the fine-tuning of large-scale models under resource constraints. △ Less

Submitted 29 May, 2026; originally announced May 2026.

Comments: KDD2026

A Non-Volatile Heterogeneous Quantum Dot III-V/Si DFB Laser with Optical Memristive Behavior

Authors: Stanley Cheung, Bassem Tossoun, Di Liang, Yuan Yuan, Yingtao Hu, Geza Kurczveil, Xucheng Yang, Raymond Beausoleil

Abstract: In this work, we introduce a non-volatile heterogeneous quantum dot (QD) III-V/Al2O3/Si distributed feedback (DFB) laser exhibiting optical memristive behavior. The device operates in the O-band (~1300 nm) with a threshold current density of 234 A/cm2 and a side-mode suppression ratio exceeding 48 dB. Co-integrated Al2O3-based memristors produce bipolar resistive switching, yielding non-volatile w… ▽ More In this work, we introduce a non-volatile heterogeneous quantum dot (QD) III-V/Al2O3/Si distributed feedback (DFB) laser exhibiting optical memristive behavior. The device operates in the O-band (~1300 nm) with a threshold current density of 234 A/cm2 and a side-mode suppression ratio exceeding 48 dB. Co-integrated Al2O3-based memristors produce bipolar resistive switching, yielding non-volatile wavelength shifts of ~ 46 pm and ~ 17 dB peak power contrast with zero static holding power. The III-V/Al2O3/Si heterojunction memristor I-V hysteresis is also modeled. This new device enables simultaneous coherent light generation and persistent optical state storage, establishing a new class of active photonic memory for neuromorphic and reconfigurable WDM applications. △ Less

Submitted 21 May, 2026; originally announced May 2026.

Comments: 13 pages, 8 figures

Viscosity Solutions of Stochastic Hamilton--Jacobi--Bellman Equations with Jumps

Authors: Dunxiang Liang, Qingxin Meng

Abstract: This paper studies the stochastic optimal control of jump-diffusion processes and the associated fully nonlinear backward stochastic Hamilton--Jacobi--Bellman (BSHJB) equations. We establish the dynamic programming principle (DPP) via backward semigroups to characterize the value function. To handle non-local integro-differential operators and polynomial growth, we introduce a stochastic viscosity… ▽ More This paper studies the stochastic optimal control of jump-diffusion processes and the associated fully nonlinear backward stochastic Hamilton--Jacobi--Bellman (BSHJB) equations. We establish the dynamic programming principle (DPP) via backward semigroups to characterize the value function. To handle non-local integro-differential operators and polynomial growth, we introduce a stochastic viscosity solution framework based on semimartingale test functions and global tangency conditions. Existence is proved using the measurable selection theorem and the generalized Itô--Kunita formula. Finally, under a super-parabolicity condition, we establish a weak comparison principle and prove global uniqueness via localized bounding envelopes and backward induction. △ Less

Submitted 19 May, 2026; originally announced May 2026.

Advancing Narrative Long Video Generation via Training-Free Identity-Aware Memory

Authors: Jinzhuo Liu, Jiangning Zhang, Wencan Jiang, Yabiao Wang, Dingkang Liang, Zhucun Xue, Ran Yi, Yong Liu

Abstract: Autoregressive video generation has improved rapidly in visual fidelity and interactivity, but it still suffers from long-term inconsistency and memory degradation. Most existing solutions either compress historical frames using predefined strategies or retrieve keyframes based on coarse implicit attention signals, both of which fail to handle evolving prompts with shifting entity references, lead… ▽ More Autoregressive video generation has improved rapidly in visual fidelity and interactivity, but it still suffers from long-term inconsistency and memory degradation. Most existing solutions either compress historical frames using predefined strategies or retrieve keyframes based on coarse implicit attention signals, both of which fail to handle evolving prompts with shifting entity references, leading to identity drift, character duplication, and attribute loss. To address this, we propose IAMFlow, a training-free identity-aware memory framework that explicitly models and tracks persistent entity identities, enabling consistent generation across prompt transitions. Specifically, an LLM extracts entities with visual attributes from each prompt and assigns unique global IDs for identity-aware memory, while a VLM asynchronously verifies and refines attributes from rendered frames, enabling explicit entity tracking in place of implicit similarity-based matching. To keep the proposed framework computationally practical, we design a systematic inference acceleration pipeline, including asynchronous visual verification, adaptive prompt transition, and model quantization, which achieves faster generation than existing baselines. Furthermore, we introduce NarraStream-Bench, a benchmark for narrative streaming video generation that features 324 multi-prompt scripts spanning six dimensions and a three-dimensional evaluation protocol that integrates both traditional metrics and multimodal large language model-based assessments. Extensive experiments show that IAMFlow, despite being training-free, achieves the best overall performance on NarraStream-Bench, outperforming the strongest baseline by 2.56 points, while achieving a 1.39$\times$ speedup over the most efficient baseline in the 60-second multi-prompt setting. △ Less

Submitted 18 May, 2026; originally announced May 2026.

Comments: Project page: https://eddie0521.github.io/projects/iamflow/ Code: https://github.com/Eddie0521/IAMFlow

Generative 3D Gaussians with Learned Density Control

Authors: Runjie Yan, Yan-Pei Cao, Peng Wang, Ding Liang, Yuan-Chen Guo

Abstract: We present Density-Sampled Gaussians (DeG), a novel 3D representation designed to bridge the gap between adaptive rendering primitives and scalable generative modeling. Unlike existing approaches that constrain 3D Gaussians to fixed voxel grids or arrays, DeG models Gaussian centers as samples from a learnable probability density function defined over an octree. This formulation provides a rigorou… ▽ More We present Density-Sampled Gaussians (DeG), a novel 3D representation designed to bridge the gap between adaptive rendering primitives and scalable generative modeling. Unlike existing approaches that constrain 3D Gaussians to fixed voxel grids or arrays, DeG models Gaussian centers as samples from a learnable probability density function defined over an octree. This formulation provides a rigorous mathematical framework for adaptive density control: by jointly optimizing the spatial density and Gaussian attributes under rendering supervision, our model naturally concentrates primitives in regions of high geometric complexity. We achieve this via a new render loss contribution gradient that serves as a fully differentiable analogue to the discrete densification and pruning heuristics used in standard Gaussian Splatting. The resulting representation is highly flexible, supporting variable-resolution decoding from a single latent code by simply adjusting the sampling budget. To enable generative synthesis, we train a latent diffusion model on DeG. We identify a critical challenge in applying diffusion to unordered set-structured latents, which can significantly slow convergence, and propose VecSeq, a canonical re-indexing mechanism that anchors latent tokens to a deterministic 3D Sobol sequence. This transforms the ambiguous set-generation problem into a robust sequence modeling task. Extensive experiments demonstrate that our pipeline achieves state-of-the-art quality in single-image-to-3D generation, combining the structural adaptivity of unstructured primitives with the training stability of grid-based methods. △ Less

Submitted 8 May, 2026; originally announced May 2026.

Comments: 19 pages, 16 figures, SIGGRAPH Conference Papers '26

ACM Class: I.4.5

Holder Policy Optimisation

Authors: Yuxiang Chen, Dingli Liang, Yihang Chen, Ziqin Gong, Chenyang Le, Zhaokai Wang, Jiachen Zhu, Lingyu Yang, Jianghao Lin, Weinan Zhang, Jun Wang

Abstract: Group Relative Policy Optimisation (GRPO) enhances large language models by estimating advantages across a group of sampled trajectories. However, mapping these trajectory-level advantages to policy updates requires aggregating token-level probabilities within each sequence. Relying on a fixed aggregation mechanism for this step fundamentally limits the algorithm's adaptability. Empirically, we ob… ▽ More Group Relative Policy Optimisation (GRPO) enhances large language models by estimating advantages across a group of sampled trajectories. However, mapping these trajectory-level advantages to policy updates requires aggregating token-level probabilities within each sequence. Relying on a fixed aggregation mechanism for this step fundamentally limits the algorithm's adaptability. Empirically, we observe a critical trade-off: certain fixed aggregations frequently suffer from training collapse, while others fail to yield satisfactory performance. To resolve this, we propose \textbf{HölderPO}, a generalised policy optimisation framework unifying token-level probability aggregation via the Hölder mean. By explicitly modulating the parameter $p$, our framework provides continuous control over the trade-off between gradient concentration and variance bounds. Theoretically, we prove that a larger $p$ concentrates the gradient to amplify sparse learning signals, whereas a smaller $p$ strictly bounds gradient variance. Because no static configuration can universally resolve this concentration-stability trade-off, we instantiate the framework with a dynamic annealing algorithm that progressively schedules $p$ across the training lifecycle. Extensive evaluations demonstrate superior stability and convergence over existing baselines. Specifically, our approach achieves a state-of-the-art average accuracy of $54.9\%$ across multiple mathematical benchmarks, yielding a substantial $7.2\%$ relative gain over standard GRPO and secures an exceptional $93.8\%$ success rate on ALFWorld. △ Less

Submitted 21 May, 2026; v1 submitted 12 May, 2026; originally announced May 2026.

Dynamically Characterizing the Structures of Dirac Points via Wave Packets

Authors: Dan-Dan Liang, Xin Shen, Zhi Li

Abstract: Topological non-trivial band structures are the core problem in the field of topological materials. In this paper, we investigate the topological band structure in a system with controllable Dirac points from the perspective of wave packet dynamics. By adding a third-nearest-neighboring coupling to the graphene model, additional pairs of Dirac points emerge. The emergence and annihilation of Dirac… ▽ More Topological non-trivial band structures are the core problem in the field of topological materials. In this paper, we investigate the topological band structure in a system with controllable Dirac points from the perspective of wave packet dynamics. By adding a third-nearest-neighboring coupling to the graphene model, additional pairs of Dirac points emerge. The emergence and annihilation of Dirac points result in hybrid and parabolic points, and we show that these band structures can be revealed by the dynamical behaviors of wave packets. Particularly, for the gapped hybrid point, the motion of the wave packet shows a one-dimensional \emph{Zitterbewegung} motion. Furthermore, we also show that the winding number associated with the Dirac point and parabolic point can be determined via the center-of-mass and spin texture of wave packets, respectively. The results of this work could motivate new experimental methods to characterize the system's topological signatures through wave packet dynamics, which may also find application in systems of other exotic topological materials. △ Less

Submitted 6 May, 2026; originally announced May 2026.

Comments: 6 pages, 5 figures. comments are welcome

Journal ref: Chinese Physics Letters 40, 110302 (2023)

Divide and Conquer: Decoupled Representation Alignment for Multimodal World Models

Authors: Junyuan Xiao, Dingkang Liang, Xin Zhou, Yixuan Ye, Tongtong Su, Guangmo Yi, Bin Xia, Qiang Lyu, Shurui Shi, Jun Huang, Jianlou Si, Wenming Yang

Abstract: Emerging multi-modal world models attempt to jointly generate videos across diverse modalities (e.g., RGB, depth, and mask), yet they fail to fully exploit the rich priors of existing foundation models. We propose $M^2$-REPA, the first representation alignment method tailored for multi-modal video generation. Our key insight is that foundation models trained on different modality spaces naturally… ▽ More Emerging multi-modal world models attempt to jointly generate videos across diverse modalities (e.g., RGB, depth, and mask), yet they fail to fully exploit the rich priors of existing foundation models. We propose $M^2$-REPA, the first representation alignment method tailored for multi-modal video generation. Our key insight is that foundation models trained on different modality spaces naturally capture distinct domain-specific priors, acting as complementary "experts." Specifically, we first decouple modality-specific features from the diffusion model's intermediate representations, then align each with its corresponding expert foundation model. To this end, we design two synergistic objectives: a multi-modal representation alignment loss that enforces feature-to-expert matching, and a modality-specific decoupling regularization that encourages complementarity across different modalities. This design enables joint optimization, fully exploiting priors from multiple foundation models. Extensive experiments demonstrate that our method significantly outperforms baselines in visual quality and long-term consistency. △ Less

Submitted 1 July, 2026; v1 submitted 3 May, 2026; originally announced May 2026.

Comments: Accepted to ECCV 2026

Beyond Semantic Relevance: Counterfactual Risk Minimization for Robust Retrieval-Augmented Generation

Authors: Peiyang Liu, Qiang Yan, Ziqiang Cui, Di Liang, Xi Wang, Wei Ye

Abstract: Standard Retrieval-Augmented Generation (RAG) systems predominantly rely on semantic relevance as a proxy for utility. However, this assumption collapses in realistic decision-making scenarios where user queries are laden with cognitive biases, such as false premises or confirmation bias. In such cases, maximizing relevance paradoxically promotes the retrieval of sycophantic evidence that reinforc… ▽ More Standard Retrieval-Augmented Generation (RAG) systems predominantly rely on semantic relevance as a proxy for utility. However, this assumption collapses in realistic decision-making scenarios where user queries are laden with cognitive biases, such as false premises or confirmation bias. In such cases, maximizing relevance paradoxically promotes the retrieval of sycophantic evidence that reinforces hallucinations, a critical failure we term the ``Relevance-Robustness Gap''. To bridge this gap, we propose CoRM-RAG (Counterfactual Risk Minimization for RAG), a framework that aligns retrieval with decision safety rather than mere similarity. Grounded in causal intervention, we introduce a Cognitive Perturbation Protocol to simulate user biases during training, which is then distilled into a lightweight Evidence Critic. This scoring module learns to identify documents that possess sufficient evidential strength to steer the model toward correctness despite adversarial query perturbations. Extensive experiments on decision-making benchmarks demonstrate that CoRM-RAG significantly outperforms strong dense retrievers and LLM-based rerankers in adversarial settings, while enabling effective risk-aware abstention through reliable robustness scoring. Our code is available at https://github.com/PeiYangLiu/CoRM-RAG.git. △ Less

Submitted 2 May, 2026; originally announced May 2026.

Chain of Evidence: Pixel-Level Visual Attribution for Iterative Retrieval-Augmented Generation

Authors: Peiyang Liu, Ziqiang Cui, Xi Wang, Di Liang, Wei Ye

Abstract: Iterative Retrieval-Augmented Generation (iRAG) has emerged as a powerful paradigm for answering complex multi-hop questions by progressively retrieving and reasoning over external documents. However, current systems predominantly operate on parsed text, which creates two critical bottlenecks: (1) \textit{Coarse-grained attribution}, where users are burdened with manually locating evidence within… ▽ More Iterative Retrieval-Augmented Generation (iRAG) has emerged as a powerful paradigm for answering complex multi-hop questions by progressively retrieving and reasoning over external documents. However, current systems predominantly operate on parsed text, which creates two critical bottlenecks: (1) \textit{Coarse-grained attribution}, where users are burdened with manually locating evidence within lengthy documents based on vague text-level citations; and (2) \textit{Visual semantic loss}, where the conversion of visually rich documents (e.g., slides, PDFs with charts) into text discards spatial logic and layout cues essential for reasoning. To bridge this gap, we present \textbf{Chain of Evidence (CoE)}, a retriever-agnostic visual attribution framework that leverages Vision-Language Models to reason directly over screenshots of retrieved document candidates. CoE eliminates format-specific parsing and outputs precise bounding boxes, visualizing the complete reasoning chain within the retrieved candidate set. We evaluate CoE on two distinct benchmarks: \textbf{Wiki-CoE}, a large-scale dataset of structured web pages derived from 2WikiMultiHopQA, and \textbf{SlideVQA}, a challenging dataset of presentation slides featuring complex diagrams and free-form layouts. Experiments demonstrate that fine-tuned Qwen3-VL-8B-Instruct achieves robust performance, significantly outperforming text-based baselines in scenarios requiring visual layout understanding, while establishing a retriever-agnostic solution for pixel-level interpretable iRAG. Our code is available at https://github.com/PeiYangLiu/CoE.git. △ Less

Submitted 23 May, 2026; v1 submitted 2 May, 2026; originally announced May 2026.

HERMES++: Toward a Unified Driving World Model for 3D Scene Understanding and Generation

Authors: Xin Zhou, Dingkang Liang, Xiwu Chen, Feiyang Tan, Dingyuan Zhang, Hengshuang Zhao, Xiang Bai

Abstract: Driving world models serve as a pivotal technology for autonomous driving by simulating environmental dynamics. However, existing approaches predominantly focus on future scene generation, often overlooking comprehensive 3D scene understanding. Conversely, while Large Language Models (LLMs) demonstrate impressive reasoning capabilities, they lack the capacity to predict future geometric evolution,… ▽ More Driving world models serve as a pivotal technology for autonomous driving by simulating environmental dynamics. However, existing approaches predominantly focus on future scene generation, often overlooking comprehensive 3D scene understanding. Conversely, while Large Language Models (LLMs) demonstrate impressive reasoning capabilities, they lack the capacity to predict future geometric evolution, creating a significant disparity between semantic interpretation and physical simulation. To bridge this gap, we propose HERMES++, a unified driving world model that integrates 3D scene understanding and future geometry prediction within a single framework. Our approach addresses the distinct requirements of these tasks through synergistic designs. First, a BEV representation consolidates multi-view spatial information into a structure compatible with LLMs. Second, we introduce LLM-enhanced world queries to facilitate knowledge transfer from the understanding branch. Third, a Current-to-Future Link is designed to bridge the temporal gap, conditioning geometric evolution on semantic context. Finally, to enforce structural integrity, we employ a Joint Geometric Optimization strategy that integrates explicit geometric constraints with implicit latent regularization to align internal representations with geometry-aware priors. Extensive evaluations on multiple benchmarks validate the effectiveness of our method. HERMES++ achieves strong performance, outperforming specialist approaches in both future point cloud prediction and 3D scene understanding tasks. The model and code will be publicly released at https://github.com/H-EmbodVis/HERMESV2. △ Less

Submitted 30 April, 2026; originally announced April 2026.

Comments: Extended version of ICCV 25 paper HERMES, Code: https://github.com/H-EmbodVis/HERMESV2, Project page: https://h-embodvis.github.io/HERMESV2/

A Deployable Embodied Vision-Language Navigation System with Hierarchical Cognition and Context-Aware Exploration

Authors: Kuan Xu, Ruimeng Liu, Yizhuo Yang, Denan Liang, Tongxing Jin, Shenghai Yuan, Chen Wang, Lihua Xie

Abstract: Bridging the gap between embodied intelligence and embedded deployment remains a key challenge in intelligent robotic systems, where perception, reasoning, and planning must operate under strict constraints on computation, memory, energy, and real-time execution. In vision-and-language navigation (VLN), existing approaches often face a trade-off between reasoning capability and deployment efficien… ▽ More Bridging the gap between embodied intelligence and embedded deployment remains a key challenge in intelligent robotic systems, where perception, reasoning, and planning must operate under strict constraints on computation, memory, energy, and real-time execution. In vision-and-language navigation (VLN), existing approaches often face a trade-off between reasoning capability and deployment efficiency on real-world platforms. In this paper, we present a deployable embodied VLN system that achieves both high efficiency and strong high-level reasoning on real-world robots. The system is decomposed into a fast perception-action layer and a deep reasoning layer running asynchronously at different time scales, with a shared memory layer enabling efficient interaction between them. To support long-horizon reasoning, we incrementally construct a compact memory graph and progressively feed decomposed subgraphs into a vision-language model (VLM). Furthermore, we formulate exploration as a Weighted Traveling Repairman Problem (WTRP) by jointly considering reasoning outcomes and the spatial distribution of candidate regions. Extensive experiments in simulation and real-world environments demonstrate improved navigation success and efficiency over existing VLN approaches while maintaining real-time performance on resource-constrained hardware. Code and additional real-world experiments are available at https://github.com/xukuanHIT/HiCo-Nav. △ Less

Submitted 16 May, 2026; v1 submitted 23 April, 2026; originally announced April 2026.

Comments: 10 pages, 5 figures,

Object Referring-Guided Scanpath Prediction with Perception-Enhanced Vision-Language Models

Authors: Rong Quan, Yantao Lai, Dong Liang, Jie Qin

Abstract: Object Referring-guided Scanpath Prediction (ORSP) aims to predict the human attention scanpath when they search for a specific target object in a visual scene according to a linguistic description describing the object. Multimodal information fusion is a key point of ORSP. Therefore, we propose a novel model, ScanVLA, to first exploit a Vision-Language Model (VLM) to extract and fuse inherently a… ▽ More Object Referring-guided Scanpath Prediction (ORSP) aims to predict the human attention scanpath when they search for a specific target object in a visual scene according to a linguistic description describing the object. Multimodal information fusion is a key point of ORSP. Therefore, we propose a novel model, ScanVLA, to first exploit a Vision-Language Model (VLM) to extract and fuse inherently aligned visual and linguistic feature representations from the input image and referring expression. Next, to enhance the ScanVLA's perception of fine-grained positional information, we not only propose a novel History Enhanced Scanpath Decoder (HESD) that directly takes historical fixations' position information as input to help predict a more reasonable position for the current fixation, but also adopt a frozen Segmentation LoRA as an auxiliary component to help localize the referred object more precisely, which improves the scanpath prediction task without incurring additional large computational and time costs. Extensive experimental results demonstrate that ScanVLA can significantly outperform existing scanpath prediction methods under object referring. △ Less

Submitted 22 April, 2026; originally announced April 2026.

Comments: ICMR 2026

Parameter Importance is Not Static: Evolving Parameter Isolation for Supervised Fine-Tuning

Authors: Zekai Lin, Chao Xue, Di Liang, Xingsheng Han, Peiyang Liu, Xianjie Wu, Lei Jiang, Yu Lu, Haibo Shi, Shuang Liang, Minlong Peng

Abstract: Supervised Fine-Tuning (SFT) of large language models often suffers from task interference and catastrophic forgetting. Recent approaches alleviate this issue by isolating task-critical parameters during training. However, these methods represent a static solution to a dynamic problem, assuming that parameter importance remains fixed once identified. In this work, we empirically demonstrate that p… ▽ More Supervised Fine-Tuning (SFT) of large language models often suffers from task interference and catastrophic forgetting. Recent approaches alleviate this issue by isolating task-critical parameters during training. However, these methods represent a static solution to a dynamic problem, assuming that parameter importance remains fixed once identified. In this work, we empirically demonstrate that parameter importance exhibits temporal drift over the course of training. To address this, we propose Evolving Parameter Isolation (EPI), a fine-tuning framework that adapts isolation decisions based on online estimates of parameter importance. Instead of freezing a fixed subset of parameters, EPI periodically updates isolation masks using gradient-based signals, enabling the model to protect emerging task-critical parameters while releasing outdated ones to recover plasticity. Experiments on diverse multi-task benchmarks demonstrate that EPI consistently reduces interference and forgetting compared to static isolation and standard fine-tuning, while improving overall generalization. Our analysis highlights the necessity of synchronizing isolation mechanisms with the evolving dynamics of learning diverse abilities. △ Less

Submitted 15 April, 2026; originally announced April 2026.

SOLARIS: Speculative Offloading of Latent-bAsed Representation for Inference Scaling

Authors: Zikun Liu, Liang Luo, Qianru Li, Zhengyu Zhang, Wei Ling, Jingyi Shen, Zeliang Chen, Yaning Huang, Jingxian Huang, Abdallah Aboelela, Chonglin Sun, Feifan Gu, Fenggang Wu, Hang Qu, Huayu Li, Jill Pan, Kaidi Pei, Laming Chen, Longhao Jin, Qin Huang, Tongyi Tang, Varna Puvvada, Wenlin Chen, Xiaohan Wei, Xu Cao , et al. (9 additional authors not shown)

Abstract: Recent advances in recommendation scaling laws have led to foundation models of unprecedented complexity. While these models offer superior performance, their computational demands make real-time serving impractical, often forcing practitioners to rely on knowledge distillation-compromising serving quality for efficiency. To address this challenge, we present SOLARIS (Speculative Offloading of Lat… ▽ More Recent advances in recommendation scaling laws have led to foundation models of unprecedented complexity. While these models offer superior performance, their computational demands make real-time serving impractical, often forcing practitioners to rely on knowledge distillation-compromising serving quality for efficiency. To address this challenge, we present SOLARIS (Speculative Offloading of Latent-bAsed Representation for Inference Scaling), a novel framework inspired by speculative decoding. SOLARIS proactively precomputes user-item interaction embeddings by predicting which user-item pairs are likely to appear in future requests, and asynchronously generating their foundation model representations ahead of time. This approach decouples the costly foundation model inference from the latency-critical serving path, enabling real-time knowledge transfer from models previously considered too expensive for online use. Deployed across Meta's advertising system serving billions of daily requests, SOLARIS achieves 0.67% revenue-driving top-line metrics gain, demonstrating its effectiveness at scale. △ Less

Submitted 3 June, 2026; v1 submitted 13 April, 2026; originally announced April 2026.

Comments: Accepted to SIGIR 2026 Industry Track

Learning from Contrasts: Synthesizing Reasoning Paths from Diverse Search Trajectories

Authors: Peiyang Liu, Zhirui Chen, Xi Wang, Di Liang, Youru Li, Zhi Cai, Wei Ye

Abstract: Monte Carlo Tree Search (MCTS) has been widely used for automated reasoning data exploration, but current supervision extraction methods remain inefficient. Standard approaches retain only the single highest-reward trajectory, discarding the comparative signals present in the many explored paths. Here we introduce \textbf{Contrastive Reasoning Path Synthesis (CRPS)}, a framework that transforms su… ▽ More Monte Carlo Tree Search (MCTS) has been widely used for automated reasoning data exploration, but current supervision extraction methods remain inefficient. Standard approaches retain only the single highest-reward trajectory, discarding the comparative signals present in the many explored paths. Here we introduce \textbf{Contrastive Reasoning Path Synthesis (CRPS)}, a framework that transforms supervision extraction from a filtering process into a synthesis procedure. CRPS uses a structured reflective process to analyze the differences between high- and low-quality search trajectories, extracting explicit information about strategic pivots and local failure modes. These insights guide the synthesis of reasoning chains that incorporate success patterns while avoiding identified pitfalls. We show empirically that models fine-tuned on just 60K CRPS-synthesized examples match or exceed the performance of baselines trained on 590K examples derived from standard rejection sampling, a 20$\times$ reduction in dataset size. Furthermore, CRPS improves generalization on out-of-domain benchmarks, demonstrating that learning from the contrast between success and failure produces more transferable reasoning capabilities than learning from success alone. △ Less

Submitted 13 April, 2026; originally announced April 2026.

Why Supervised Fine-Tuning Fails to Learn: A Systematic Study of Incomplete Learning in Large Language Models

Authors: Chao Xue, Yao Wang, Mengqiao Liu, Di Liang, Xingsheng Han, Peiyang Liu, Xianjie Wu, Chenyao Lu, Lei Jiang, Yu Lu, Haibo Shi, Shuang Liang, Minlong Peng, Flora D. Salim

Abstract: Supervised Fine-Tuning (SFT) is the standard approach for adapting large language models (LLMs) to downstream tasks. However, we observe a persistent failure mode: even after convergence, models often fail to correctly reproduce a subset of their own supervised training data. We refer to this behavior as the Incomplete Learning Phenomenon(ILP). This paper presents the first systematic study of ILP… ▽ More Supervised Fine-Tuning (SFT) is the standard approach for adapting large language models (LLMs) to downstream tasks. However, we observe a persistent failure mode: even after convergence, models often fail to correctly reproduce a subset of their own supervised training data. We refer to this behavior as the Incomplete Learning Phenomenon(ILP). This paper presents the first systematic study of ILP in LLM fine-tuning. We formalize ILP as post-training failure to internalize supervised instances and demonstrate its prevalence across multiple model families, domains, and datasets. Through controlled analyses, we identify five recurrent sources of incomplete learning: (1) missing prerequisite knowledge in the pre-trained model, (2) conflicts between SFT supervision and pre-training knowledge, (3) internal inconsistencies within SFT data, (4) left-side forgetting during sequential fine-tuning, and (5) insufficient optimization for rare or complex patterns. We introduce a diagnostic-first framework that maps unlearned samples to these causes using observable training and inference signals, and study several targeted mitigation strategies as causal interventions. Experiments on Qwen, LLaMA, and OLMo2 show that incomplete learning is widespread and heterogeneous, and that improvements in aggregate metrics can mask persistent unlearned subsets. The findings highlight the need for fine-grained diagnosis of what supervised fine-tuning fails to learn, and why. △ Less

Submitted 24 April, 2026; v1 submitted 11 April, 2026; originally announced April 2026.

Comments: Accepted by ACL 2026 Main

Reason Only When Needed: Efficient Generative Reward Modeling via Model-Internal Uncertainty

Authors: Chao Xue, Yao Wang, Mengqiao Liu, Di Liang, Xingsheng Han, Peiyang Liu, Xianjie Wu, Chenyao Lu, Lei Jiang, Yu Lu, Haibo Shi, Shuang Liang, Minlong Peng, Flora D. Salim

Abstract: Recent advancements in the Generative Reward Model (GRM) have demonstrated its potential to enhance the reasoning abilities of LLMs through Chain-of-Thought (CoT) prompting. Despite these gains, existing implementations of GRM suffer from two critical limitations. First, CoT prompting is applied indiscriminately to all inputs regardless of their inherent complexity. This introduces unnecessary com… ▽ More Recent advancements in the Generative Reward Model (GRM) have demonstrated its potential to enhance the reasoning abilities of LLMs through Chain-of-Thought (CoT) prompting. Despite these gains, existing implementations of GRM suffer from two critical limitations. First, CoT prompting is applied indiscriminately to all inputs regardless of their inherent complexity. This introduces unnecessary computational costs for tasks amenable to fast, direct inference. Second, existing approaches primarily rely on voting-based mechanisms to evaluate CoT outputs, which often lack granularity and precision in assessing reasoning quality. In this paper, we propose E-GRM, an efficient generative reward modeling framework grounded in model-internal uncertainty. E-GRM leverages the convergence behavior of parallel model generations to estimate uncertainty and selectively trigger CoT reasoning only when needed, without relying on handcrafted features or task-dependent signals. To improve reward fidelity, we introduce a lightweight discriminative scorer trained with a hybrid regression--ranking objective to provide fine-grained evaluation of reasoning paths. Experiments on multiple reasoning benchmarks show that E-GRM substantially reduces inference cost while consistently improving answer accuracy, demonstrating that model-internal uncertainty is an effective and general signal for efficient reasoning-aware reward modeling. △ Less

Submitted 3 May, 2026; v1 submitted 11 April, 2026; originally announced April 2026.

Comments: accepted by ACL 2026

When Numbers Speak: Aligning Textual Numerals and Visual Instances in Text-to-Video Diffusion Models

Authors: Zhengyang Sun, Yu Chen, Xin Zhou, Xiaofan Li, Xiwu Chen, Dingkang Liang, Xiang Bai

Abstract: Text-to-video diffusion models have enabled open-ended video synthesis, but often struggle with generating the correct number of objects specified in a prompt. We introduce NUMINA , a training-free identify-then-guide framework for improved numerical alignment. NUMINA identifies prompt-layout inconsistencies by selecting discriminative self- and cross-attention heads to derive a countable latent l… ▽ More Text-to-video diffusion models have enabled open-ended video synthesis, but often struggle with generating the correct number of objects specified in a prompt. We introduce NUMINA , a training-free identify-then-guide framework for improved numerical alignment. NUMINA identifies prompt-layout inconsistencies by selecting discriminative self- and cross-attention heads to derive a countable latent layout. It then refines this layout conservatively and modulates cross-attention to guide regeneration. On the introduced CountBench, NUMINA improves counting accuracy by up to 7.4% on Wan2.1-1.3B, and by 4.9% and 5.5% on 5B and 14B models, respectively. Furthermore, CLIP alignment is improved while maintaining temporal consistency. These results demonstrate that structural guidance complements seed search and prompt enhancement, offering a practical path toward count-accurate text-to-video diffusion. The code is available at https://github.com/H-EmbodVis/NUMINA. △ Less

Submitted 9 April, 2026; originally announced April 2026.

Comments: Accepted by CVPR 2026. Project page: https://h-embodvis.github.io/NUMINA

Coalitional Zero-Sum Games for ${H_{\infty}}$ Leader-Following Consensus Control

Authors: Yunxiao Ren, Dingguo Liang, Yuezu Lv, Zhisheng Duan

Abstract: This paper investigates the leader-following consensus problem for a class of multi-agent systems subject to adversarial attack-like external inputs. To address this, we formulate the robust leader-following control problem as a global coalitional min-max zero-sum game using differential game theory. Specifically, the agents' control inputs form a coalition to minimize a global cost function, whil… ▽ More This paper investigates the leader-following consensus problem for a class of multi-agent systems subject to adversarial attack-like external inputs. To address this, we formulate the robust leader-following control problem as a global coalitional min-max zero-sum game using differential game theory. Specifically, the agents' control inputs form a coalition to minimize a global cost function, while the attacks form an opposing coalition to maximize it. Notably, when these external adversarial attacks manifest as disturbances, the designed game-theoretic control policy systematically yields a robust $H_\infty$ control law. Addressing this problem inherently requires solving a high-dimensional generalized algebraic Riccati equation (GARE), which poses significant challenges for distributed computation and controller implementation. To overcome these challenges, we propose a two-fold approach. First, a decentralized computational strategy is devised to decompose the high-dimensional GARE into multiple uniform, lower-dimensional GAREs. Second, a dynamic average consensus-based decoupling algorithm is developed to resolve the inherent coupling structure of the robust control law, thereby facilitating its distributed implementation. Finally, numerical simulations on the formation control of multi-vehicle systems with feedback-linearized dynamics are conducted to validate the effectiveness of the proposed algorithms. △ Less

Submitted 7 April, 2026; originally announced April 2026.