Showing 1–50 of 1,080 results for author: Lin, W

$π$-Play: Multi-Agent Self-Play via Privileged Self-Distillation without External Data

Authors: Yaocheng Zhang, Yuanheng Zhu, Wenyue Chong, Songjun Tu, Qichao Zhang, Jiajun Chai, Xiaohan Wang, Wei Lin, Guojun Yin, Dongbin Zhao

Abstract: Deep search agents have emerged as a promising paradigm for addressing complex information-seeking tasks, but their training remains challenging due to sparse rewards, weak credit assignment, and limited labeled data. Self-play offers a scalable route to reduce data dependence, but conventional self-play optimizes students only through sparse outcome rewards, leading to low learning efficiency. In… ▽ More Deep search agents have emerged as a promising paradigm for addressing complex information-seeking tasks, but their training remains challenging due to sparse rewards, weak credit assignment, and limited labeled data. Self-play offers a scalable route to reduce data dependence, but conventional self-play optimizes students only through sparse outcome rewards, leading to low learning efficiency. In this work, we observe that self-play naturally produces a question construction path (QCP) during task generation, an intermediate artifact that captures the reverse solution process. This reveals a new source of privileged information for self-distillation: self-play can itself provide high-quality privileged context for the teacher model in a low-cost and scalable manner, without relying on human feedback or curated privileged information. Leveraging this insight, we propose Privileged Information Self-Play ($π$-Play), a multi-agent self-evolution framework. In $π$-Play, an examiner generates tasks together with their QCPs, and a teacher model leverages QCP as privileged context to densely supervise a student via self-distillation. This design transforms conventional sparse-reward self-play into a dense-feedback self-evolution loop. Extensive experiments show that data-free $π$-Play surpasses fully supervised search agents and improves evolutionary efficiency by 2-3$\times$ over conventional self-play. △ Less

Submitted 15 April, 2026; originally announced April 2026.

Comments: 26 pages, 12 figures

Throughput Characterization of Wireless CSMA Networks With Arbitrary Sensing and Interference Topologies

Authors: Xinghua Sun, Wenhai Lin, Ruike Zhou

Abstract: The performance analysis of wireless CSMA networks is notoriously difficult due to the intricate sensing and interference relationships among links. Even the fundamental problem of throughput characterization remains open when sensing and interference topologies are both arbitrary. In this paper, we develop a new analytical framework for throughput characterization in wireless CSMA networks with a… ▽ More The performance analysis of wireless CSMA networks is notoriously difficult due to the intricate sensing and interference relationships among links. Even the fundamental problem of throughput characterization remains open when sensing and interference topologies are both arbitrary. In this paper, we develop a new analytical framework for throughput characterization in wireless CSMA networks with arbitrary sensing and interference topologies. The proposed framework yields explicit throughput expressions without relying on the commonly adopted zero-propagation-delay assumption. The key idea is to exploit the clique structure of the sensing graph to transform the original CSMA network into an equivalent multi-channel network, and then model its dynamics through a discrete-time Markov renewal process. In this way, the framework explicitly captures global coupling among links and enables analytical evaluation of how access parameters affect network performance. The proposed analysis is applied to several representative CSMA scenarios, including networks with multi-BSS IEEE 802.11 networks with universal frequency reuse, and ad-hoc topologies exhibiting hidden-terminal, exposed-terminal, and flow-in-the-middle effects. Simulation results show that, in dense deployments and in scenarios with strong coupling among link behaviors, the proposed model significantly outperforms existing analytical approaches in throughput estimation and enables more accurate determination of access parameters. △ Less

Submitted 14 April, 2026; originally announced April 2026.

Bottleneck Tokens for Unified Multimodal Retrieval

Authors: Siyu Sun, Jing Ren, Zhaohe Liao, Dongxiao Mao, Xiangyuan Ren, Yiyi Zhang, Haohua Zhao, Weixiong Lin, Jiang Shaohua, Liqing Zhang, Yuchao Zheng

Abstract: Adapting decoder-only multimodal large language models (MLLMs) for unified multimodal retrieval faces two structural gaps. First, existing methods rely on implicit pooling, which overloads the hidden state of a standard vocabulary token (e.g., <EOS>) as the sequence-level representation, a mechanism never designed for information aggregation. Second, contrastive fine-tuning specifies what the embe… ▽ More Adapting decoder-only multimodal large language models (MLLMs) for unified multimodal retrieval faces two structural gaps. First, existing methods rely on implicit pooling, which overloads the hidden state of a standard vocabulary token (e.g., <EOS>) as the sequence-level representation, a mechanism never designed for information aggregation. Second, contrastive fine-tuning specifies what the embedding should match but provides no token-level guidance on how information should be compressed into it. We address both gaps with two complementary components. Architecturally, we introduce Bottleneck Tokens (BToks), a small set of learnable tokens that serve as a fixed-capacity explicit pooling mechanism. For training, we propose Generative Information Condensation: a next-token prediction objective coupled with a Condensation Mask that severs the direct attention path from target tokens to query tokens. All predictive signals are thereby forced through the BToks, converting the generative loss into dense, token-level supervision for semantic compression. At inference time, only the input and BToks are processed in a single forward pass with negligible overhead over conventional last-token pooling. On MMEB-V2 (78 datasets, 3 modalities, 9 meta-tasks), our approach achieves state-of-the-art among 2B-scale methods under comparable data conditions, attaining an Overall score of 59.0 (+3.6 over VLM2Vec-V2) with substantial gains on semantically demanding tasks (e.g., +12.6 on Video-QA). △ Less

Submitted 13 April, 2026; originally announced April 2026.

Mem$^2$Evolve: Towards Self-Evolving Agents via Co-Evolutionary Capability Expansion and Experience Distillation

Authors: Zihao Cheng, Zeming Liu, Yingyu Shan, Xinyi Wang, Xiangrong Zhu, Yunpu Ma, Hongru Wang, Yuhang Guo, Wei Lin, Yunhong Wang

Abstract: While large language model--powered agents can self-evolve by accumulating experience or by dynamically creating new assets (i.e., tools or expert agents), existing frameworks typically treat these two evolutionary processes in isolation. This separation overlooks their intrinsic interdependence: the former is inherently bounded by a manually predefined static toolset, while the latter generates n… ▽ More While large language model--powered agents can self-evolve by accumulating experience or by dynamically creating new assets (i.e., tools or expert agents), existing frameworks typically treat these two evolutionary processes in isolation. This separation overlooks their intrinsic interdependence: the former is inherently bounded by a manually predefined static toolset, while the latter generates new assets from scratch without experiential guidance, leading to limited capability growth and unstable evolution. To address this limitation, we introduce a novel paradigm of co-evolutionary Capability Expansion and Experience Distillation. Guided by this paradigm, we propose the \textbf{Mem$^{\textbf{2}}$Evolve}, which integrates two core components: \textbf{Experience Memory} and \textbf{Asset Memory}. Specifically, Mem$^{2}$Evolve leverages accumulated experience to guide the dynamic creation of assets, thereby expanding the agent's capability space while simultaneously acquiring new experience to achieve co-evolution. Extensive experiments across 6 task categories and 8 benchmarks demonstrate that Mem$^{2}$Evolve achieves improvement of 18.53\% over standard LLMs, 11.80\% over agents evolving solely through experience, and 6.46\% over those evolving solely through asset creation, establishing it as a substantially more effective and stable self-evolving agent framework. Code is available at: https://buaa-irip-llm.github.io/Mem2Evolve. △ Less

Submitted 12 April, 2026; originally announced April 2026.

Comments: Accepted by ACL 2026 Main

DiningBench: A Hierarchical Multi-view Benchmark for Perception and Reasoning in the Dietary Domain

Authors: Song Jin, Juntian Zhang, Xun Zhang, Zeying Tian, Fei Jiang, Guojun Yin, Wei Lin, Yong Liu, Rui Yan

Abstract: Recent advancements in Vision-Language Models (VLMs) have revolutionized general visual understanding. However, their application in the food domain remains constrained by benchmarks that rely on coarse-grained categories, single-view imagery, and inaccurate metadata. To bridge this gap, we introduce DiningBench, a hierarchical, multi-view benchmark designed to evaluate VLMs across three levels of… ▽ More Recent advancements in Vision-Language Models (VLMs) have revolutionized general visual understanding. However, their application in the food domain remains constrained by benchmarks that rely on coarse-grained categories, single-view imagery, and inaccurate metadata. To bridge this gap, we introduce DiningBench, a hierarchical, multi-view benchmark designed to evaluate VLMs across three levels of cognitive complexity: Fine-Grained Classification, Nutrition Estimation, and Visual Question Answering. Unlike previous datasets, DiningBench comprises 3,021 distinct dishes with an average of 5.27 images per entry, incorporating fine-grained "hard" negatives from identical menus and rigorous, verification-based nutritional data. We conduct an extensive evaluation of 29 state-of-the-art open-source and proprietary models. Our experiments reveal that while current VLMs excel at general reasoning, they struggle significantly with fine-grained visual discrimination and precise nutritional reasoning. Furthermore, we systematically investigate the impact of multi-view inputs and Chain-of-Thought reasoning, identifying five primary failure modes. DiningBench serves as a challenging testbed to drive the next generation of food-centric VLM research. All codes are released in https://github.com/meituan/DiningBench. △ Less

Submitted 11 April, 2026; originally announced April 2026.

Comments: ACL 2026 Main

Mild Over-Parameterization Benefits Asymmetric Tensor PCA

Authors: Shihong Ding, Weicheng Lin, Cong Fang

Abstract: Asymmetric Tensor PCA (ATPCA) is a prototypical model for studying the trade-offs between sample complexity, computation, and memory. Existing algorithms for this problem typically require at least $d^{\left\lceil\overline{k}/2\right\rceil}$ state memory cost to recover the signal, where $d$ is the vector dimension and $\overline{k}$ is the tensor order. We focus on the setting where… ▽ More Asymmetric Tensor PCA (ATPCA) is a prototypical model for studying the trade-offs between sample complexity, computation, and memory. Existing algorithms for this problem typically require at least $d^{\left\lceil\overline{k}/2\right\rceil}$ state memory cost to recover the signal, where $d$ is the vector dimension and $\overline{k}$ is the tensor order. We focus on the setting where $\overline{k} \geq 4$ is even and consider (stochastic) gradient descent-based algorithms under a limited memory budget, which permits only mild over-parameterization of the model. We propose a matrix-parameterized method (in $d^{2}$ state memory cost) using a novel three-phase alternating-update algorithm to address the problem and demonstrate how mild over-parameterization facilitates learning in two key aspects: (i) it improves sample efficiency, allowing our method to achieve \emph{near-optimal} $d^{\overline{k}-2}$ sample complexity in our limited memory setting; and (ii) it enhances adaptivity to problem structure, a previously unrecognized phenomenon, where the required sample size naturally decreases as consecutive vectors become more aligned, and in the symmetric limit attains $d^{\overline{k}/2}$, matching the \emph{best} known polynomial-time complexity. To our knowledge, this is the \emph{first} tractable algorithm for ATPCA with $d^{\overline{k}}$-independent memory costs. △ Less

Submitted 11 April, 2026; originally announced April 2026.

U$^{2}$Flow: Uncertainty-Aware Unsupervised Optical Flow Estimation

Authors: Xunpei Sun, Wenwei Lin, Yi Chang, Gang Chen

Abstract: Unsupervised optical flow methods typically lack reliable uncertainty estimation, limiting their robustness and interpretability. We propose U$^{2}$Flow, the first recurrent unsupervised framework that jointly estimates optical flow and per-pixel uncertainty. The core innovation is a decoupled learning strategy that derives uncertainty supervision from augmentation consistency via a Laplace-based… ▽ More Unsupervised optical flow methods typically lack reliable uncertainty estimation, limiting their robustness and interpretability. We propose U$^{2}$Flow, the first recurrent unsupervised framework that jointly estimates optical flow and per-pixel uncertainty. The core innovation is a decoupled learning strategy that derives uncertainty supervision from augmentation consistency via a Laplace-based maximum likelihood objective, enabling stable training without ground truth. The predicted uncertainty is further integrated into the network to guide adaptive flow refinement and dynamically modulate the regional smoothness loss. Furthermore, we introduce an uncertainty-guided bidirectional flow fusion mechanism that enhances robustness in challenging regions. Extensive experiments on KITTI and Sintel demonstrate that U$^{2}$Flow achieves state-of-the-art performance among unsupervised methods while producing highly reliable uncertainty maps, validating the effectiveness of our joint estimation paradigm. The code is available at https://github.com/sunzunyi/U2FLOW. △ Less

Submitted 11 April, 2026; originally announced April 2026.

Comments: Accepted as an oral presentation at CVPR 2026

Multi-task Just Recognizable Difference for Video Coding for Machines: Database, Model, and Coding Application

Authors: Junqi Liu, Yun Zhang, Xiaoxia Huang, Long Xu, Weisi Lin

Abstract: Just Recognizable Difference (JRD) boosts coding efficiency for machine vision through visibility threshold modeling, but is currently limited to a single-task scenario. To address this issue, we propose a Multi-Task JRD (MT-JRD) dataset and an Attribute-assisted MT-JRD (AMT-JRD) model for Video Coding for Machines (VCM), enhancing both prediction accuracy and coding efficiency. First, we construc… ▽ More Just Recognizable Difference (JRD) boosts coding efficiency for machine vision through visibility threshold modeling, but is currently limited to a single-task scenario. To address this issue, we propose a Multi-Task JRD (MT-JRD) dataset and an Attribute-assisted MT-JRD (AMT-JRD) model for Video Coding for Machines (VCM), enhancing both prediction accuracy and coding efficiency. First, we construct a dataset comprising 27,264 JRD annotations from machines, supporting three representative tasks including object detection, instance segmentation, and keypoint detection. Secondly, we propose the AMT-JRD prediction model, which integrates Generalized Feature Extraction Module (GFEM) and Specialized Feature Extraction Module (SFEM) to facilitate joint learning across multiple tasks. Thirdly, we innovatively incorporate object attribute information into object-wise JRD prediction through the Attribute Feature Fusion Module (AFFM), which introduces prior knowledge about object size and location. This design effectively compensates for the limitations of relying solely on image features and enhances the model's capacity to represent the perceptual mechanisms of machine vision. Finally, we apply the AMT-JRD model to VCM, where the accurately predicted JRDs are applied to reduce the coding bit rate while preserving accuracy across multiple machine vision tasks. Extensive experimental results demonstrate that AMT-JRD achieves precise and robust multi-task prediction with a mean absolute error of 3.781 and error variance of 5.332 across three tasks, outperforming the state-of-the-art single-task prediction model by 6.7% and 6.3%, respectively. Coding experiments further reveal that compared to the baseline VVC and JPEG, the AMT-JRD-based VCM improves an average of 3.861% and 7.886% Bjontegaard Delta-mean Average Precision (BD-mAP), respectively. △ Less

Submitted 10 April, 2026; originally announced April 2026.

Comments: Submitted to IEEE Transactions on Circuits and Systems for Video Technology

GPAFormer: Graph-guided Patch Aggregation Transformer for Efficient 3D Medical Image Segmentation

Authors: Chung-Ming Lo, I-Yun Liu, Wei-Yang Lin

Abstract: Deep learning has been widely applied to 3D medical image segmentation tasks. However, due to the diversity of imaging modalities, the high-dimensional nature of the data, and the heterogeneity of anatomical structures, achieving both segmentation accuracy and computational efficiency in multi-organ segmentation remains a challenge. This study proposed GPAFormer, a lightweight network architecture… ▽ More Deep learning has been widely applied to 3D medical image segmentation tasks. However, due to the diversity of imaging modalities, the high-dimensional nature of the data, and the heterogeneity of anatomical structures, achieving both segmentation accuracy and computational efficiency in multi-organ segmentation remains a challenge. This study proposed GPAFormer, a lightweight network architecture specifically designed for 3D medical image segmentation, emphasizing efficiency while keeping high accuracy. GPAFormer incorporated two core modules: the multi-scale attention-guided stacked aggregation (MASA) and the mutual-aware patch graph aggregator (MPGA). MASA utilized three parallel paths with different receptive fields, combined through planar aggregation, to enhance the network's capability in handling structures of varying sizes. MPGA employed a graph-guided approach to dynamically aggregate regions with similar feature distributions based on inter-patch feature similarity and spatial adjacency, thereby improving the discrimination of both internal and boundary structures of organs. Experiments were performed on public whole-body CT and MRI datasets including BTCV, Synapse, ACDC, and BraTS. Compared to the existed 3D segmentation networkd, GPAFormer using only 1.81 M parameters achieved overall highest DSC on BTCV (75.70%), Synapse (81.20%), ACDC (89.32%), and BraTS (82.74%). Using consumer level GPU, the inference time for one validation case of BTCV spent less than one second. The results demonstrated that GPAFormer balanced accuracy and efficiency in multi-organ, multi-modality 3D segmentation tasks across various clinical scenarios especially for resource-constrained and time-sensitive clinical environments. △ Less

Submitted 8 April, 2026; originally announced April 2026.

CI-ICM: Channel Importance-driven Learned Image Coding for Machines

Authors: Yun Zhang, Junle Liu, Huan Zhang, Zhaoqing Pan, Gangyi Jiang, Weisi Lin

Abstract: Traditional human vision-centric image compression methods are suboptimal for machine vision centric compression due to different visual properties and feature characteristics. To address this problem, we propose a Channel Importance-driven learned Image Coding for Machines (CI-ICM), aiming to maximize the performance of machine vision tasks at a given bitrate constraint. First, we propose a Chann… ▽ More Traditional human vision-centric image compression methods are suboptimal for machine vision centric compression due to different visual properties and feature characteristics. To address this problem, we propose a Channel Importance-driven learned Image Coding for Machines (CI-ICM), aiming to maximize the performance of machine vision tasks at a given bitrate constraint. First, we propose a Channel Importance Generation (CIG) module to quantify channel importance in machine vision and develop a channel order loss to rank channels in descending order. Second, to properly allocate bitrate among feature channels, we propose a Feature Channel Grouping and Scaling (FCGS) module that non-uniformly groups the feature channels based on their importance and adjusts the dynamic range of each group. Based on FCGS, we further propose a Channel Importance-based Context (CI-CTX) module to allocate bits among feature groups and to preserve higher fidelity in critical channels. Third, to adapt to multiple machine tasks, we propose a Task-Specific Channel Adaptation (TSCA) module to adaptively enhance features for multiple downstream machine tasks. Experimental results on the COCO2017 dataset show that the proposed CI-ICM achieves BD-mAP@50:95 gains of 16.25$\%$ in object detection and 13.72$\%$ in instance segmentation over the established baseline codec. Ablation studies validate the effectiveness of each contribution, and computation complexity analysis reveals the practicability of the CI-ICM. This work establishes feature channel optimization for machine vision-centric compression, bridging the gap between image coding and machine perception. △ Less

Submitted 6 April, 2026; originally announced April 2026.

NTIRE 2026 3D Restoration and Reconstruction in Real-world Adverse Conditions: RealX3D Challenge Results

Authors: Shuhong Liu, Chenyu Bao, Ziteng Cui, Xuangeng Chu, Bin Ren, Lin Gu, Xiang Chen, Mingrui Li, Long Ma, Marcos V. Conde, Radu Timofte, Yun Liu, Ryo Umagami, Tomohiro Hashimoto, Zijian Hu, Yuan Gan, Tianhan Xu, Yusuke Kurose, Tatsuya Harada, Junwei Yuan, Gengjia Chang, Xining Ge, Mache You, Qida Cao, Zeliang Li , et al. (81 additional authors not shown)

Abstract: This paper presents a comprehensive review of the NTIRE 2026 3D Restoration and Reconstruction (3DRR) Challenge, detailing the proposed methods and results. The challenge seeks to identify robust reconstruction pipelines that are robust under real-world adverse conditions, specifically extreme low-light and smoke-degraded environments, as captured by our RealX3D benchmark. A total of 279 participa… ▽ More This paper presents a comprehensive review of the NTIRE 2026 3D Restoration and Reconstruction (3DRR) Challenge, detailing the proposed methods and results. The challenge seeks to identify robust reconstruction pipelines that are robust under real-world adverse conditions, specifically extreme low-light and smoke-degraded environments, as captured by our RealX3D benchmark. A total of 279 participants registered for the competition, of whom 33 teams submitted valid results. We thoroughly evaluate the submitted approaches against state-of-the-art baselines, revealing significant progress in 3D reconstruction under adverse conditions. Our analysis highlights shared design principles among top-performing methods and provides insights into effective strategies for handling 3D scene degradation. △ Less

Submitted 5 April, 2026; originally announced April 2026.

Predicting Dynamics of Ultra-Large Complex Systems by Inferring Governing Equations

Authors: Qi Shao, Duxin Chen, Jiawen Chen, Yujie Zeng, Athen Ma, Wenwu Yu, Vito Latora, Wei Lin

Abstract: Predicting the behavior of ultra-large complex systems, from climate to biological and technological networks, is a central unsolved challenge. Existing approaches face a fundamental trade-off: equation discovery methods provide interpretability but fail to scale, while neural networks scale but operate as black boxes and often lose reliability over long times. Here, we introduce the Sparse Identi… ▽ More Predicting the behavior of ultra-large complex systems, from climate to biological and technological networks, is a central unsolved challenge. Existing approaches face a fundamental trade-off: equation discovery methods provide interpretability but fail to scale, while neural networks scale but operate as black boxes and often lose reliability over long times. Here, we introduce the Sparse Identification Graph Neural Network, a framework that overcome this divide by allowing to infer the governing equations of large networked systems from data. By defining symbolic discovery as edge-level information, SIGN decouples the scalability of sparse identification from network size, enabling efficient equation discovery even in large systems. SIGN allows to study networks with over 100,000 nodes while remaining robust to noise, sparse sampling, and missing data. Across diverse benchmark systems, including coupled chaotic oscillators, neural dynamics, and epidemic spreading, it recovers governing equations with high precision and sustains accurate long-term predictions. Applied to a data set of time series of temperature measurements in 71,987 sea surface positions, SIGN identifies a compact predictive network model and captures large-scale sea surface temperature conditions up to two years in advance. By enabling equation discovery at previously inaccessible scales, SIGN opens a path toward interpretable and reliable prediction of real-world complex systems. △ Less

Submitted 1 April, 2026; originally announced April 2026.

Comments: 15 pages, 5 figures, under review

GenoBERT: A Language Model for Accurate Genotype Imputation

Authors: Lei Huang, Chuan Qiu, Kuan-Jui Su, Anqi Liu, Yun Gong, Weiqiang Lin, Lindong Jiang, Chen Zhao, Meng Song, Jeffrey Deng, Qing Tian, Zhe Luo, Ping Gong, Hui Shen, Chaoyang Zhang, Hong-Wen Deng

Abstract: Genotype imputation enables dense variant coverage for genome-wide association and risk-prediction studies, yet conventional reference-panel methods remain limited by ancestry bias and reduced rare-variant accuracy. We present Genotype Bidirectional Encoder Representations from Transformers (GenoBERT), a transformer-based, reference-free framework that tokenizes phased genotypes and uses a self-at… ▽ More Genotype imputation enables dense variant coverage for genome-wide association and risk-prediction studies, yet conventional reference-panel methods remain limited by ancestry bias and reduced rare-variant accuracy. We present Genotype Bidirectional Encoder Representations from Transformers (GenoBERT), a transformer-based, reference-free framework that tokenizes phased genotypes and uses a self-attention mechanism to capture both short- and long-range linkage disequilibrium (LD) dependencies. Benchmarking on two independent datasets including the Louisiana Osteoporosis Study (LOS) and the 1000 Genomes Project (1KGP) across ancestry groups and multiple genotype missingness levels (5-50%) shows that GenoBERT achieves the highest overall accuracy compared to four baseline methods (Beagle5.4, SCDA, BiU-Net, and STICI). At practical sparsity levels (up to 25% missing), GenoBERT attains high overall imputation accuracy ($r^2 approx 0.98$) across datasets, and maintains robust performance ($r^2 > 0.90$) even at 50% missingness. Experimental results across different ancestries confirm consistent gains across datasets, with resilience to small sample sizes and weak LD. A 128-SNP (single-nucleotide polymorphism) context window (approximately 100 Kb) is validated through LD-decay analyses as sufficient to capture local correlation structures. By eliminating reference-panel dependence while preserving high accuracy, GenoBERT provides a scalable and robust solution for genotype imputation and a foundation for downstream genomic modeling. △ Less

Submitted 31 March, 2026; originally announced April 2026.

Criterion Validity of LLM-as-Judge for Business Outcomes in Conversational Commerce

Authors: Liang Chen, Qi Liu, Wenhuan Lin, Feng Liang

Abstract: Multi-dimensional rubric-based dialogue evaluation is widely used to assess conversational AI, yet its criterion validity -- whether quality scores are associated with the downstream outcomes they are meant to serve -- remains largely untested. We address this gap through a two-phase study on a major Chinese matchmaking platform, testing a 7-dimension evaluation rubric (implemented via LLM-as-Judg… ▽ More Multi-dimensional rubric-based dialogue evaluation is widely used to assess conversational AI, yet its criterion validity -- whether quality scores are associated with the downstream outcomes they are meant to serve -- remains largely untested. We address this gap through a two-phase study on a major Chinese matchmaking platform, testing a 7-dimension evaluation rubric (implemented via LLM-as-Judge) against verified business conversion. Our findings concern rubric design and weighting, not LLM scoring accuracy: any judge using the same rubric would face the same structural issue. The core finding is dimension-level heterogeneity: in Phase 2 (n=60 human conversations, stratified sample, verified labels), Need Elicitation (D1: rho=0.368, p=0.004) and Pacing Strategy (D3: rho=0.354, p=0.006) are significantly associated with conversion after Bonferroni correction, while Contextual Memory (D5: rho=0.018, n.s.) shows no detectable association. This heterogeneity causes the equal-weighted composite (rho=0.272) to underperform its best dimensions -- a composite dilution effect that conversion-informed reweighting partially corrects (rho=0.351). Logistic regression controlling for conversation length confirms D3's association strengthens (OR=3.18, p=0.006), ruling out a length confound. An initial pilot (n=14) mixing human and AI conversations had produced a misleading "evaluation-outcome paradox," which Phase 2 revealed as an agent-type confound artifact. Behavioral analysis of 130 conversations through a Trust-Funnel framework identifies a candidate mechanism: AI agents execute sales behaviors without building user trust. We operationalize these findings in a three-layer evaluation architecture and advocate criterion validity testing as standard practice in applied dialogue evaluation. △ Less

Submitted 11 March, 2026; originally announced April 2026.

DipGuava: Disentangling Personalized Gaussian Features for 3D Head Avatars from Monocular Video

Authors: Jeonghaeng Lee, Seok Keun Choi, Zhixuan Li, Weisi Lin, Sanghoon Lee

Abstract: While recent 3D head avatar creation methods attempt to animate facial dynamics, they often fail to capture personalized details, limiting realism and expressiveness. To fill this gap, we present DipGuava (Disentangled and Personalized Gaussian UV Avatar), a novel 3D Gaussian head avatar creation method that successfully generates avatars with personalized attributes from monocular video. DipGuava… ▽ More While recent 3D head avatar creation methods attempt to animate facial dynamics, they often fail to capture personalized details, limiting realism and expressiveness. To fill this gap, we present DipGuava (Disentangled and Personalized Gaussian UV Avatar), a novel 3D Gaussian head avatar creation method that successfully generates avatars with personalized attributes from monocular video. DipGuava is the first method to explicitly disentangle facial appearance into two complementary components, trained in a structured two-stage pipeline that significantly reduces learning ambiguity and enhances reconstruction fidelity. In the first stage, we learn a stable geometry-driven base appearance that captures global facial structure and coarse expression-dependent variations. In the second stage, the personalized residual details not captured in the first stage are predicted, including high-frequency components and nonlinearly varying features such as wrinkles and subtle skin deformations. These components are fused via dynamic appearance fusion that integrates residual details after deformation, ensuring spatial and semantic alignment. This disentangled design enables DipGuava to generate photorealistic, identity-preserving avatars, consistently outperforming prior methods in both visual quality and quantitativeperformance, as demonstrated in extensive experiments. △ Less

Submitted 29 March, 2026; originally announced March 2026.

Comments: AAAI 2026

PhySkin: Physics-based Bone-driven Neural Garment Simulation

Authors: Astitva Srivastava, Hsiao-yu Chen, Ryan Goldade, Philipp Herholz, Zhongshi Jiang, Gene Wei-Chin Lin, Lingchen Yang, Nikolaos Sarafianos, Tuur Stuyck, Egor Larionov

Abstract: Recent advances in digital avatar technology have enabled the generation of compelling virtual characters, but deploying these avatars on compute-constrained devices poses significant challenges for achieving realistic garment deformations. While physics-based simulations yield accurate results, they are computationally prohibitive for real-time applications. Conversely, linear blend skinning offe… ▽ More Recent advances in digital avatar technology have enabled the generation of compelling virtual characters, but deploying these avatars on compute-constrained devices poses significant challenges for achieving realistic garment deformations. While physics-based simulations yield accurate results, they are computationally prohibitive for real-time applications. Conversely, linear blend skinning offers efficiency but fails to capture the complex dynamics of loose-fitting garments, resulting in unrealistic motion and visual artifacts. Neural methods have shown promise, yet they struggle to animate loose clothing plausibly under strict performance constraints. In this work, we present a novel approach for fast and physically plausible garment draping tailored for resource-constrained environments. Our method leverages a reduced-space quasi-static neural simulation, mapping the garment's full degrees of freedom to a set of bone handles that drive deformation. A neural deformation model is trained in a fully self-supervised manner, eliminating the need for costly simulation data. At runtime, a lightweight neural network modulates the handle deformations based on body shape and pose, enabling realistic garment behavior that respects physical properties such as gravity, fabric stretching, bending, and collision avoidance. Experimental results demonstrate that our method achieves physically plausible garment drapes while generalizing across diverse poses and body shapes, supporting zero-shot evaluation and mesh topology independence. Our method's runtime significantly outperforms past works, as it runs in microseconds per frame using single-threaded CPU inference, offering a practical solution for real-time avatar animation on low-compute devices. △ Less

Submitted 27 March, 2026; originally announced March 2026.

AgentCollab: A Self-Evaluation-Driven Collaboration Paradigm for Efficient LLM Agents

Authors: Wenbo Gao, Renxi Liu, Xian Wang, Fang Guo, Shuai Yang, Xi Chen, Hui-Ling Zhen, Hanting Chen, Weizhe Lin, Xiaosong Li, Yaoyuan Wang

Abstract: Autonomous agents powered by large language models (LLMs) perform complex tasks through long-horizon reasoning and tool interaction, where a fundamental trade-off arises between execution efficiency and reasoning robustness. Models at different capability-cost levels offer complementary advantages: lower-cost models enable fast execution but may struggle on difficult reasoning segments, while stro… ▽ More Autonomous agents powered by large language models (LLMs) perform complex tasks through long-horizon reasoning and tool interaction, where a fundamental trade-off arises between execution efficiency and reasoning robustness. Models at different capability-cost levels offer complementary advantages: lower-cost models enable fast execution but may struggle on difficult reasoning segments, while stronger models provide more robust reasoning at higher computational cost. We present AgentCollab, a self-driven collaborative inference framework that dynamically coordinates models with different reasoning capacities during agent execution. Instead of relying on external routing modules, the framework uses the agent's own self-reflection signal to determine whether the current reasoning trajectory is making meaningful progress, and escalates control to a stronger reasoning tier only when necessary. To further stabilize long-horizon execution, we introduce a difficulty-aware cumulative escalation strategy that allocates additional reasoning budget based on recent failure signals. In our experiments, we instantiate this framework using a two-level small-large model setting. Experiments on diverse multi-step agent benchmarks show that AgentCollab consistently improves the accuracy-efficiency Pareto frontier of LLM agents. △ Less

Submitted 26 March, 2026; originally announced March 2026.

Composer 2 Technical Report

Authors: Cursor Research, :, Aaron Chan, Ahmed Shalaby, Alexander Wettig, Aman Sanger, Andrew Zhai, Anurag Ajay, Ashvin Nair, Charlie Snell, Chen Lu, Chen Shen, Emily Jia, Federico Cassano, Hanpeng Liu, Haoyu Chen, Henry Wildermuth, Jacob Jackson, Janet Li, Jediah Katz, Jiajun Yao, Joey Hejna, Josh Warner, Julius Vering, Kevin Frans , et al. (31 additional authors not shown)

Abstract: Composer 2 is a specialized model designed for agentic software engineering. The model demonstrates strong long-term planning and coding intelligence while maintaining the ability to efficiently solve problems for interactive use. The model is trained in two phases: first, continued pretraining to improve the model's knowledge and latent coding ability, followed by large-scale reinforcement learni… ▽ More Composer 2 is a specialized model designed for agentic software engineering. The model demonstrates strong long-term planning and coding intelligence while maintaining the ability to efficiently solve problems for interactive use. The model is trained in two phases: first, continued pretraining to improve the model's knowledge and latent coding ability, followed by large-scale reinforcement learning to improve end-to-end coding performance through stronger reasoning, accurate multi-step execution, and coherence on long-horizon realistic coding problems. We develop infrastructure to support training in the same Cursor harness that is used by the deployed model, with equivalent tools and structure, and use environments that match real problems closely. To measure the ability of the model on increasingly difficult tasks, we introduce a benchmark derived from real software engineering problems in large codebases including our own. Composer 2 is a frontier-level coding model and demonstrates a process for training strong domain-specialized models. On our CursorBench evaluations the model achieves a major improvement in accuracy compared to previous Composer models (61.3). On public benchmarks the model scores 61.7 on Terminal-Bench and 73.7 on SWE-bench Multilingual in our harness, comparable to state-of-the-art systems. △ Less

Submitted 25 March, 2026; v1 submitted 25 March, 2026; originally announced March 2026.

MemDLM: Memory-Enhanced DLM Training

Authors: Zehua Pei, Hui-Ling Zhen, Weizhe Lin, Sinno Jialin Pan, Yunhe Wang, Mingxuan Yuan, Bei Yu

Abstract: Diffusion Language Models (DLMs) offer attractive advantages over Auto-Regressive (AR) models, such as full-attention parallel decoding and flexible generation. However, standard DLM training uses a static, single-step masked prediction objective that never exposes the model to the progressive denoising dynamics of inference, and forces all contextual information to be maintained purely through to… ▽ More Diffusion Language Models (DLMs) offer attractive advantages over Auto-Regressive (AR) models, such as full-attention parallel decoding and flexible generation. However, standard DLM training uses a static, single-step masked prediction objective that never exposes the model to the progressive denoising dynamics of inference, and forces all contextual information to be maintained purely through token-space attention, which becomes increasingly diluted as context length grows. We propose MemDLM (Memory-Enhanced DLM), which introduces a second memory channel by embedding a simulated denoising trajectory into training via Bi-level Optimization. An inner loop updates a set of fast weights, forming a Parametric Memory that captures the local trajectory experience, while an outer loop updates the base model conditioned on this memory. By offloading part of the memorization burden from token-space attention to parameter space, MemDLM yields faster convergence, stronger long-context representations, and lower training loss, even when the fast weights are discarded at inference time. Re-enabling the inner loop at inference provides an additional prompt-specific adaptation effect, where the Parametric Memory acts as an emergent in-weight retrieval mechanism on challenging Needle-in-a-Haystack tasks. Code: https://github.com/JarvisPei/MemDLM. △ Less

Submitted 13 April, 2026; v1 submitted 23 March, 2026; originally announced March 2026.

The data heat island effect: quantifying the impact of AI data centers in a warming world

Authors: Andrea Marinoni, Erik Cambria, Luca Dal Zilio, Weisi Lin, Mauro Dalla Mura, Jocelyn Chanussot, Edoardo Ragusa, Chi Yan Tso, Yihao Zhu, Benjamin Horton

Abstract: The strong and continuous increase of AI-based services leads to the steady proliferation of AI data centres worldwide with the unavoidable escalation of their power consumption. It is unknown how this energy demand for computational purposes will impact the surrounding environment. Here, we focus our attention on the heat dissipation of AI hyperscalers. Taking advantage of land surface temperatur… ▽ More The strong and continuous increase of AI-based services leads to the steady proliferation of AI data centres worldwide with the unavoidable escalation of their power consumption. It is unknown how this energy demand for computational purposes will impact the surrounding environment. Here, we focus our attention on the heat dissipation of AI hyperscalers. Taking advantage of land surface temperature measurements acquired by remote sensing platforms over the last decades, we are able to obtain a robust assessment of the temperature increase recorded in the areas surrounding AI data centres globally. We estimate that the land surface temperature increases by 2°C on average after the start of operations of an AI data centre, inducing local microclimate zones, which we call the data heat island effect. We assess the impact on the communities, quantifying that more than 340 million people could be affected by this temperature increase. Our results show that the data heat island effect could have a remarkable influence on communities and regional welfare in the future, hence becoming part of the conversation around environmentally sustainable AI worldwide. △ Less

Submitted 1 April, 2026; v1 submitted 21 March, 2026; originally announced March 2026.

VEPO: Variable Entropy Policy Optimization for Low-Resource Language Foundation Models

Authors: Chonghan Liu, Yimin Du, Qi An, Xin He, Cunqi Zhai, Fei Tan, Weijia Lin, Xiaochun Gong, Yongchao Deng, Shousheng Jia, Xiangzheng Zhang

Abstract: Large language models frequently exhibit suboptimal performance on low resource languages, primarily due to inefficient subword segmentation and systemic training data imbalances. In this paper, we propose Variable Entropy Policy Optimization (VEPO), which leverages Reinforcement Learning with Verifiable Rewards to incorporate deterministic structural constraints into the policy alignment process.… ▽ More Large language models frequently exhibit suboptimal performance on low resource languages, primarily due to inefficient subword segmentation and systemic training data imbalances. In this paper, we propose Variable Entropy Policy Optimization (VEPO), which leverages Reinforcement Learning with Verifiable Rewards to incorporate deterministic structural constraints into the policy alignment process. This framework ensures prescribed sequence length, robust format consistency, and rigorous linguistic well formedness, all enforced during training. Central to our approach is a variable entropy mechanism that enables the model to dynamically calibrate the equilibrium between literal fidelity and semantic naturalness by modulating the exploration exploitation manifold. By integrating entropy tempered advantage estimation with asymmetric clipping, VEPO sustains robust exploration while mitigating policy collapse. Empirical evaluations across 90 FLORES-200, COMET-22, chrF directions demonstrate that VEPO yields substantial improvements in both tokenization efficiency and translation quality, bridging the performance gap for underrepresented languages. △ Less

Submitted 19 March, 2026; originally announced March 2026.

Comments: 23 pages. Includes figures and tables. Conference submission

GenLie: A Global-Enhanced Lie Detection Network under Sparsity and Semantic Interference

Authors: Zongshun Zhang, Yao Liu, Qiao Liu, Xuefeng Peng, Peiyuan Jiang, Jiaye Yang, Daibing Yao, Wei Lin

Abstract: Video-based lie detection aims to identify deceptive behaviors from visual cues. Despite recent progress, its core challenge lies in learning sparse yet discriminative representations. Deceptive signals are typically subtle and short-lived, easily overwhelmed by redundant information, while individual and contextual variations introduce strong identity-related noise. To address this issue, we prop… ▽ More Video-based lie detection aims to identify deceptive behaviors from visual cues. Despite recent progress, its core challenge lies in learning sparse yet discriminative representations. Deceptive signals are typically subtle and short-lived, easily overwhelmed by redundant information, while individual and contextual variations introduce strong identity-related noise. To address this issue, we propose GenLie, a Global-Enhanced Lie Detection Network that performs local feature modeling under global supervision. Specifically, sparse and subtle deceptive cues are captured at the local level, while global supervision and optimization ensure robust and discriminative representations by suppressing identity-related noise. Experiments on three public datasets, covering both high- and low-stakes scenarios, show that GenLie consistently outperforms state-of-the-art methods. Source code is available at https://github.com/AliasDictusZ1/GenLie. △ Less

Submitted 14 March, 2026; originally announced March 2026.

Comments: Accepted to IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) 2026

Exclusivity-Guided Mask Learning for Semi-Supervised Crowd Instance Segmentation and Counting

Authors: Jiyang Huang, Hongru Cheng, Wei Lin, Jia Wan, Antoni B. Chan

Abstract: Semi-supervised crowd analysis is a prominent area of research, as unlabeled data are typically abundant and inexpensive to obtain. However, traditional point-based annotations constrain performance because individual regions are inherently ambiguous, and consequently, learning fine-grained structural semantics from sparse anno tations remains an unresolved challenge. In this paper, we first propo… ▽ More Semi-supervised crowd analysis is a prominent area of research, as unlabeled data are typically abundant and inexpensive to obtain. However, traditional point-based annotations constrain performance because individual regions are inherently ambiguous, and consequently, learning fine-grained structural semantics from sparse anno tations remains an unresolved challenge. In this paper, we first propose an Exclusion-Constrained Dual-Prompt SAM (EDP-SAM), based on our Nearest Neighbor Exclusion Circle (NNEC) constraint, to generate mask supervision for current datasets. With the aim of segmenting individuals in dense scenes, we then propose Exclusivity-Guided Mask Learning (XMask), which enforces spatial separation through a discriminative mask objective. Gaussian smoothing and a differentiable center sampling strategy are utilized to improve feature continuity and training stability. Building on XMask, we present a semi-supervised crowd counting framework that uses instance mask priors as pseudo-labels, which contain richer shape information than traditional point cues. Extensive experiments on the ShanghaiTech A, UCF-QNRF, and JHU++ datasets (using 5%, 10%, and 40% labeled data) verify that our end-to-end model achieves state-of-the-art semi-supervised segmentation and counting performance, effectively bridging the gap between counting and instance segmentation within a unified framework. △ Less

Submitted 17 March, 2026; originally announced March 2026.

GT-PCQA: Geometry-Texture Decoupled Point Cloud Quality Assessment with MLLM

Authors: Guohua Zhang, Jian Jin, Meiqin Liu, Chao Yao, Weisi Lin, Yao Zhao

Abstract: With the rapid advancement of Multi-modal Large Language Models (MLLMs), MLLM-based Image Quality Assessment (IQA) methods have shown promising generalization. However, directly extending these MLLM-based IQA methods to PCQA remains challenging. On the one hand, existing PCQA datasets are limited in scale, which hinders stable and effective instruction tuning of MLLMs. On the other hand, due to la… ▽ More With the rapid advancement of Multi-modal Large Language Models (MLLMs), MLLM-based Image Quality Assessment (IQA) methods have shown promising generalization. However, directly extending these MLLM-based IQA methods to PCQA remains challenging. On the one hand, existing PCQA datasets are limited in scale, which hinders stable and effective instruction tuning of MLLMs. On the other hand, due to large-scale image-text pretraining, MLLMs tend to rely on texture-dominant reasoning and are insufficiently sensitive to geometric structural degradations that are critical for PCQA. To address these gaps, we propose a novel MLLM-based no-reference PCQA framework, termed GT-PCQA, which is built upon two key strategies. First, to enable stable and effective instruction tuning under scarce PCQA supervision, a 2D-3D joint training strategy is proposed. This strategy formulates PCQA as a relative quality comparison problem to unify large-scale IQA datasets with limited PCQA datasets. It incorporates a parameter-efficient Low-Rank Adaptation (LoRA) scheme to support instruction tuning. Second, a geometry-texture decoupling strategy is presented, which integrates a dual-prompt mechanism with an alternating optimization scheme to mitigate the inherent texture-dominant bias of pre-trained MLLMs, while enhancing sensitivity to geometric structural degradations. Extensive experiments demonstrate that GT-PCQA achieves competitive performance and exhibits strong generalization. △ Less

Submitted 16 March, 2026; originally announced March 2026.

Ultra-Early Prediction of Tipping Points: Integrating Dynamical Measures with Reservoir Computing

Authors: Xin Li, Qunxi Zhu, Chengli Zhao, Bolin Zhao, Xue Zhang, Xiaojun Duan, Wei Lin

Abstract: Complex dynamical systems-such as climate, ecosystems, and economics-can undergo catastrophic and potentially irreversible regime changes, often triggered by environmental parameter drift and stochastic disturbances. These critical thresholds, known as tipping points, pose a prediction problem of both theoretical and practical significance, yet remain largely unresolved. To address this, we articu… ▽ More Complex dynamical systems-such as climate, ecosystems, and economics-can undergo catastrophic and potentially irreversible regime changes, often triggered by environmental parameter drift and stochastic disturbances. These critical thresholds, known as tipping points, pose a prediction problem of both theoretical and practical significance, yet remain largely unresolved. To address this, we articulate a model-free framework that integrates the measures characterizing the stability and sensitivity of dynamical systems with the reservoir computing (RC), a lightweight machine learning technique, using only observational time series data. The framework consists of two stages. The first stage involves using RC to robustly learn local complex dynamics from observational data segmented into windows. The second stage focuses on accurately detecting early warning signals of tipping points by analyzing the learned autonomous RC dynamics through dynamical measures, including the dominant eigenvalue of the Jacobian matrix, the maximum Floquet multiplier, and the maximum Lyapunov exponent. Furthermore, when these dynamical measures exhibit trend-like patterns, their extrapolation enables ultra-early prediction of tipping points significantly prior to the occurrence of critical transitions. We conduct a rigorous theoretical analysis of the proposed method and perform extensive numerical evaluations on a series of representative synthetic systems and eight real-world datasets, as well as quantitatively predict the tipping time of the Atlantic Meridional Overturning Circulation system. Experimental results demonstrate that our framework exhibits advantages over the baselines in comprehensive evaluations, particularly in terms of dynamical interpretability, prediction stability and robustness, and ultra-early prediction capability. △ Less

Submitted 16 March, 2026; originally announced March 2026.

Fine-tuning is Not Enough: A Parallel Framework for Collaborative Imitation and Reinforcement Learning in End-to-end Autonomous Driving

Authors: Zhexi Lian, Haoran Wang, Xuerun Yan, Weimeng Lin, Xianhong Zhang, Yongyu Chen, Jia Hu

Abstract: End-to-end autonomous driving is typically built upon imitation learning (IL), yet its performance is constrained by the quality of human demonstrations. To overcome this limitation, recent methods incorporate reinforcement learning (RL) through sequential fine-tuning. However, such a paradigm remains suboptimal: sequential RL fine-tuning can introduce policy drift and often leads to a performance… ▽ More End-to-end autonomous driving is typically built upon imitation learning (IL), yet its performance is constrained by the quality of human demonstrations. To overcome this limitation, recent methods incorporate reinforcement learning (RL) through sequential fine-tuning. However, such a paradigm remains suboptimal: sequential RL fine-tuning can introduce policy drift and often leads to a performance ceiling due to its dependence on the pretrained IL policy. To address these issues, we propose PaIR-Drive, a general Parallel framework for collaborative Imitation and Reinforcement learning in end-to-end autonomous driving. During training, PaIR-Drive separates IL and RL into two parallel branches with conflict-free training objectives, enabling fully collaborative optimization. This design eliminates the need to retrain RL when applying a new IL policy. During inference, RL leverages the IL policy to further optimize the final plan, allowing performance beyond prior knowledge of IL. Furthermore, we introduce a tree-structured trajectory neural sampler to group relative policy optimization (GRPO) in the RL branch, which enhances exploration capability. Extensive analysis on NAVSIMv1 and v2 benchmark demonstrates that PaIR-Drive achieves Competitive performance of 91.2 PDMS and 87.9 EPDMS, building upon Transfuser and DiffusionDrive IL baselines. PaIR-Drive consistently outperforms existing RL fine-tuning methods, and could even correct human experts' suboptimal behaviors. Qualitative results further confirm that PaIR-Drive can effectively explore and generate high-quality trajectories. △ Less

Submitted 9 April, 2026; v1 submitted 14 March, 2026; originally announced March 2026.

Comments: 11 pages, 7 figures, 6 tables

UniCom: Unified Multimodal Modeling via Compressed Continuous Semantic Representations

Authors: Yaqi Zhao, Wang Lin, Zijian Zhang, Miles Yang, Jingyuan Chen, Wentao Zhang, Zhao Zhong, Liefeng Bo

Abstract: Current unified multimodal models typically rely on discrete visual tokenizers to bridge the modality gap. However, discretization inevitably discards fine-grained semantic information, leading to suboptimal performance in visual understanding tasks. Conversely, directly modeling continuous semantic representations (e.g., CLIP, SigLIP) poses significant challenges in high-dimensional generative mo… ▽ More Current unified multimodal models typically rely on discrete visual tokenizers to bridge the modality gap. However, discretization inevitably discards fine-grained semantic information, leading to suboptimal performance in visual understanding tasks. Conversely, directly modeling continuous semantic representations (e.g., CLIP, SigLIP) poses significant challenges in high-dimensional generative modeling, resulting in slow convergence and training instability. To resolve this dilemma, we introduce UniCom, a unified framework that harmonizes multimodal understanding and generation via compressed continuous representation. We empirically demonstrate that reducing channel dimension is significantly more effective than spatial downsampling for both reconstruction and generation. Accordingly, we design an attention-based semantic compressor to distill dense features into a compact unified representation. Furthermore, we validate that the transfusion architecture surpasses query-based designs in convergence and consistency. Experiments demonstrate that UniCom achieves state-of-the-art generation performance among unified models. Notably, by preserving rich semantic priors, it delivers exceptional controllability in image editing and maintains image consistency even without relying on VAE. △ Less

Submitted 11 March, 2026; originally announced March 2026.

R4-CGQA: Retrieval-based Vision Language Models for Computer Graphics Image Quality Assessment

Authors: Zhuangzi Li, Jian Jin, Shilv Cai, Weisi Lin

Abstract: Immersive Computer Graphics (CGs) rendering has become ubiquitous in modern daily life. However, comprehensively evaluating CG quality remains challenging for two reasons: First, existing CG datasets lack systematic descriptions of rendering quality; and second existing CG quality assessment methods cannot provide reasonable text-based explanations. To address these issues, we first identify six k… ▽ More Immersive Computer Graphics (CGs) rendering has become ubiquitous in modern daily life. However, comprehensively evaluating CG quality remains challenging for two reasons: First, existing CG datasets lack systematic descriptions of rendering quality; and second existing CG quality assessment methods cannot provide reasonable text-based explanations. To address these issues, we first identify six key perceptual dimensions of CG quality from the user perspective and construct a dataset of 3500 CG images with corresponding quality descriptions. Each description covers CG style, content, and perceived quality along the selected dimensions. Furthermore, we use a subset of the dataset to build several question-answer benchmarks based on the descriptions in order to evaluate the responses of existing Vision Language Models (VLMs). We find that current VLMs are not sufficiently accurate in judging fine-grained CG quality, but that descriptions of visually similar images can significantly improve a VLM's understanding of a given CG image. Motivated by this observation, we adopt retrieval-augmented generation and propose a two-stream retrieval framework that effectively enhances the CG quality assessment capabilities of VLMs. Experiments on several representative VLMs demonstrate that our method substantially improves their performance on CG quality assessment. △ Less

Submitted 11 March, 2026; originally announced March 2026.

TPIFM: A Task-Aware Model for Evaluating Perceptual Interaction Fluency in Remote AR Collaboration

Authors: Jiarun Song, Ninghao Wan, Fuzheng Yang, Weisi Lin

Abstract: Remote Collaborative Augmented Reality (RCAR) enables geographically distributed users to collaborate by integrating virtual and physical environments. However, because RCAR relies on real-time transmission, it is susceptible to delay and stalling impairments under constrained network conditions. Perceptual interaction fluency (PIF), defined as the perceived pace and responsiveness of collaboratio… ▽ More Remote Collaborative Augmented Reality (RCAR) enables geographically distributed users to collaborate by integrating virtual and physical environments. However, because RCAR relies on real-time transmission, it is susceptible to delay and stalling impairments under constrained network conditions. Perceptual interaction fluency (PIF), defined as the perceived pace and responsiveness of collaboration, is influenced not only by physical network impairments but also by intrinsic task characteristics. These characteristics can be interpreted as the task-specific just-noticeable difference (JND), i.e., the maximal tolerable temporal responsiveness before PIF degrades. When the average response time (ART), measured as the mean time per operation from receiving collaborator feedback to initiating the next action, falls within the JND, PIF is generally sustained, whereas values exceeding it indicate disruption. Tasks differ in their JNDs, reflecting distinct temporal responsiveness demands and sensitivities to impairments. From the perspective of the Free Energy Principle (FEP), tasks with lower JNDs impose stricter temporal prediction demands, making PIF more vulnerable to impairments, whereas higher JNDs allow greater tolerance. On this basis, we classify RCAR tasks by JND and evaluate their PIF through controlled subjective experiments under delay, stalling, and hybrid conditions. Building on these findings, we propose the Task-Aware Perceptual Interaction Fluency Model (TPIFM). Experimental results show that TPIFM accurately assesses PIF under network impairments, providing guidance for adaptive RCAR design and user experience optimization under network constraints. △ Less

Submitted 10 March, 2026; originally announced March 2026.

From Perception to Cognition: How Latency Affects Interaction Fluency and Social Presence in VR Conferencing

Authors: Jiarun Song, Ninghao Wan, FuZheng Yang, Weisi Lin

Abstract: Virtual reality (VR) conferencing has the potential to provide geographically dispersed users with an immersive environment, enabling rich social interactions and user experience using avatars. However, remote communication in VR inevitably introduces end-to-end (E2E) latency, which can significantly impact user experience. To clarify the impact of latency, we conducted subjective experiments to a… ▽ More Virtual reality (VR) conferencing has the potential to provide geographically dispersed users with an immersive environment, enabling rich social interactions and user experience using avatars. However, remote communication in VR inevitably introduces end-to-end (E2E) latency, which can significantly impact user experience. To clarify the impact of latency, we conducted subjective experiments to analyze how it influences interaction fluency from the perspective of quality perception and social presence from the perspective of social cognition, comparing VR conferencing with traditional video conferencing (VC). Specifically, interaction fluency emphasizes user perception of interaction pace and responsiveness and is assessed using Absolute Category Rating (ACR) method. In contrast, social presence focuses on the cognitive understanding of interaction, specifically whether individuals can comprehend the intentions, emotions, and behaviors expressed by others. It is primarily measured using the Networked Minds Social Presence Inventory (NMSPI). Building on this analysis, we further investigate the relationship between interaction fluency and social presence under different latency conditions to clarify the underlying perceptual and cognitive mechanisms. The findings from these subjective tests provide meaningful insights for optimizing the related systems, helping to improve interaction fluency and enhancing social presence in immersive virtual environments. △ Less

Submitted 10 March, 2026; originally announced March 2026.

CDRRM: Contrast-Driven Rubric Generation for Reliable and Interpretable Reward Modeling

Authors: Dengcan Liu, Fengkai Yang, Xiaohan Wang, Shurui Yan, Jiajun Chai, Jiahao Li, Yikun Ban, Zhendong Mao, Wei Lin, Guojun Yin

Abstract: Reward modeling is essential for aligning Large Language Models(LLMs) with human preferences, yet conventional reward models suffer from poor interpretability and heavy reliance on costly expert annotations. While recent rubric-based approaches enhance evaluation transparency, they lack systematic quality control, yielding noisy and redundant criteria, failing to mitigate persistent biases (e.g.,… ▽ More Reward modeling is essential for aligning Large Language Models(LLMs) with human preferences, yet conventional reward models suffer from poor interpretability and heavy reliance on costly expert annotations. While recent rubric-based approaches enhance evaluation transparency, they lack systematic quality control, yielding noisy and redundant criteria, failing to mitigate persistent biases (e.g., verbosity, position) in LLM evaluators, and creating a scalability-reliability trade-off. To address these limitations, we propose CDRRM (Contrast-Driven Rubric Reward Model), a framework built on a novel Contrast-then-Synthesis paradigm for high-quality rubric generation and guided preference judgment. CDRRM first conducts multi-dimensional contrastive profiling on preference pairs to identify causal discriminative factors, then synthesizes these insights into compact, context-aware rubrics to guide preference judg- ments. Extensive experiments on three authoritative benchmarks (RewardBench, RMBench, RMB) demonstrate that CDRRM achieves state-of-the-art performance across diverse domains and effectively mitigates aforementioned evaluation biases. Notably, our approach delivers exceptional data efficiency: training the rubric generator on only 3k high-quality samples empowers a frozen pre-trained judge model to outperform fully fine-tuned baselines. This work offers a scalable, interpretable, and data-efficient path for reward modeling. △ Less

Submitted 9 March, 2026; originally announced March 2026.

DreamSAC: Learning Hamiltonian World Models via Symmetry Exploration

Authors: Jinzhou Tang, Fan Feng, Minghao Fu, Wenjun Lin, Biwei Huang, Keze Wang

Abstract: Learned world models excel at interpolative generalization but fail at extrapolative generalization to novel physical properties. This limitation arises because they learn statistical correlations rather than the environment's underlying generative rules, such as physical invariances and conservation laws. We argue that learning these invariances is key to robust extrapolation. To achieve this, we… ▽ More Learned world models excel at interpolative generalization but fail at extrapolative generalization to novel physical properties. This limitation arises because they learn statistical correlations rather than the environment's underlying generative rules, such as physical invariances and conservation laws. We argue that learning these invariances is key to robust extrapolation. To achieve this, we first introduce \textbf{Symmetry Exploration}, an unsupervised exploration strategy where an agent is intrinsically motivated by a Hamiltonian-based curiosity bonus to actively probe and challenge its understanding of conservation laws, thereby collecting physically informative data. Second, we design a Hamiltonian-based world model that learns from the collected data, using a novel self-supervised contrastive objective to identify the invariant physical state from raw, view-dependent pixel observations. Our framework, \textbf{DreamSAC}, trained on this actively curated data, significantly outperforms state-of-the-art baselines in 3D physics simulations on tasks requiring extrapolation. △ Less

Submitted 8 March, 2026; originally announced March 2026.

Comments: 19 pages, 5 figures

RPG-SAM: Reliability-Weighted Prototypes and Geometric Adaptive Threshold Selection for Training-Free One-Shot Polyp Segmentation

Authors: Weikun Lin, Yunhao Bai, Yan Wang

Abstract: Training-free one-shot segmentation offers a scalable alternative to expert annotations where knowledge is often transferred from support images and foundation models. But existing methods often treat all pixels in support images and query response intensities models in a homogeneous way. They ignore the regional heterogeity in support images and response heterogeity in query.To resolve this, we p… ▽ More Training-free one-shot segmentation offers a scalable alternative to expert annotations where knowledge is often transferred from support images and foundation models. But existing methods often treat all pixels in support images and query response intensities models in a homogeneous way. They ignore the regional heterogeity in support images and response heterogeity in query.To resolve this, we propose RPG-SAM, a framework that systematically tackles these heterogeneity gaps. Specifically, to address regional heterogeneity, we introduce Reliability-Weighted Prototype Mining (RWPM) to prioritize high-fidelity support features while utilizing background anchors as contrastive references for noise suppression. To address response heterogeneity, we develop Geometric Adaptive Selection (GAS) to dynamically recalibrate binarization thresholds by evaluating the morphological consensus of candidates. Finally, an iterative refinement loop method is designed to polishes anatomical boundaries. By accounting for multi-layered information heterogeneity, RPG-SAM achieves a 5.56\% mIoU improvement on the Kvasir dataset. Code will be released. △ Less

Submitted 13 April, 2026; v1 submitted 7 March, 2026; originally announced March 2026.

Comments: 8 pages, 3 figures

Rethinking Personalization in Large Language Models at the Token Level

Authors: Chenheng Zhang, Yijun Lu, Lizhe Fang, Chunyuan Zheng, Jiajun Chai, Xiaohan Wang, Guojun Yin, Wei Lin, Yisen Wang, Zhouchen Lin

Abstract: With large language models (LLMs) now performing strongly across diverse tasks, there is growing demand for them to personalize outputs for individual users. Personalization is typically framed as an additional layer on top of a base NLP task, requiring model responses to meet user-specific needs while still accomplishing the underlying task. From a token-level perspective, different tokens in a r… ▽ More With large language models (LLMs) now performing strongly across diverse tasks, there is growing demand for them to personalize outputs for individual users. Personalization is typically framed as an additional layer on top of a base NLP task, requiring model responses to meet user-specific needs while still accomplishing the underlying task. From a token-level perspective, different tokens in a response contribute to personalization to varying degrees. Tokens with higher personalization relevance should therefore receive greater emphasis when developing personalized LLMs. However, accurately estimating such personalization degrees remains challenging. To address this challenge, we propose PerContrast, a self-contrast method that estimates each output token's dependence on user-specific information through causal intervention. Building on this mechanism, we develop the PerCE loss, which adaptively upweights tokens with higher estimated personalization degrees during training via a bootstrap procedure, enabling the model to alternate between estimating and optimizing these tokens. Experiments on multiple LLMs demonstrate that PerCE substantially improves personalization performance with minimal additional cost, achieving average gains of over 10% and up to 68.04% on the LongLaMP dataset, along with strong cross-task and cross-scenario transferability. These results highlight the importance of token-level personalization modeling and establish token-aware training as a simple yet effective paradigm for advancing personalized LLMs. △ Less

Submitted 4 February, 2026; originally announced March 2026.

Skeleton-to-Image Encoding: Enabling Skeleton Representation Learning via Vision-Pretrained Models

Authors: Siyuan Yang, Jun Liu, Hao Cheng, Chong Wang, Shijian Lu, Hedvig Kjellstrom, Weisi Lin, Alex C. Kot

Abstract: Recent advances in large-scale pretrained vision models have demonstrated impressive capabilities across a wide range of downstream tasks, including cross-modal and multi-modal scenarios. However, their direct application to 3D human skeleton data remains challenging due to fundamental differences in data format. Moreover, the scarcity of large-scale skeleton datasets and the need to incorporate s… ▽ More Recent advances in large-scale pretrained vision models have demonstrated impressive capabilities across a wide range of downstream tasks, including cross-modal and multi-modal scenarios. However, their direct application to 3D human skeleton data remains challenging due to fundamental differences in data format. Moreover, the scarcity of large-scale skeleton datasets and the need to incorporate skeleton data into multi-modal action recognition without introducing additional model branches present significant research opportunities. To address these challenges, we introduce Skeleton-to-Image Encoding (S2I), a novel representation that transforms skeleton sequences into image-like data by partitioning and arranging joints based on body-part semantics and resizing to standardized image dimensions. This encoding enables, for the first time, the use of powerful vision-pretrained models for self-supervised skeleton representation learning, effectively transferring rich visual-domain knowledge to skeleton analysis. While existing skeleton methods often design models tailored to specific, homogeneous skeleton formats, they overlook the structural heterogeneity that naturally arises from diverse data sources. In contrast, our S2I representation offers a unified image-like format that naturally accommodates heterogeneous skeleton data. Extensive experiments on NTU-60, NTU-120, and PKU-MMD demonstrate the effectiveness and generalizability of our method for self-supervised skeleton representation learning, including under challenging cross-format evaluation settings. △ Less

Submitted 6 March, 2026; originally announced March 2026.

Comments: Submitted to IEEE TPAMI, under review

LocalSUG: Geography-Aware LLM for Query Suggestion in Local-Life Services

Authors: Jinwen Chen, Shuai Gong, Shiwen Zhang, Zheng Zhang, Yachao Zhao, Lingxiang Wang, Haibo Zhou, Yuan Zhan, Wei Lin, Hainan Zhang

Abstract: In local-life service platforms, the query suggestion module plays a crucial role in enhancing user experience by generating candidate queries based on user input prefixes, thus reducing user effort and accelerating search. Traditional multi-stage cascading systems rely heavily on historical top queries, limiting their ability to address long-tail demand. While LLMs offer strong semantic generaliz… ▽ More In local-life service platforms, the query suggestion module plays a crucial role in enhancing user experience by generating candidate queries based on user input prefixes, thus reducing user effort and accelerating search. Traditional multi-stage cascading systems rely heavily on historical top queries, limiting their ability to address long-tail demand. While LLMs offer strong semantic generalization, deploying them in local-life services introduces three key challenges: lack of geographic grounding, exposure bias in preference optimization, and online inference latency. To address these issues, we propose LocalSUG, an LLM-based query suggestion framework tailored for local-life service platforms. First, we introduce a city-aware candidate mining strategy based on term co-occurrence to inject geographic grounding into generation. Second, we propose a beam-search-driven GRPO algorithm that aligns training with inference-time decoding, reducing exposure bias in autoregressive generation. A multi-objective reward mechanism further optimizes both relevance and business-oriented metrics. Finally, we develop quality-aware beam acceleration and vocabulary pruning techniques that significantly reduce online latency while preserving generation quality. Extensive offline evaluations and large-scale online A/B testing demonstrate that LocalSUG improves click-through rate (CTR) by +0.35% and reduces the low/no-result rate by 2.56%, validating its effectiveness in real-world deployment. △ Less

Submitted 5 March, 2026; originally announced March 2026.

EVMbench: Evaluating AI Agents on Smart Contract Security

Authors: Justin Wang, Andreas Bigger, Xiaohai Xu, Justin W. Lin, Andy Applebaum, Tejal Patwardhan, Alpin Yukseloglu, Olivia Watkins

Abstract: Smart contracts on public blockchains now manage large amounts of value, and vulnerabilities in these systems can lead to substantial losses. As AI agents become more capable at reading, writing, and running code, it is natural to ask how well they can already navigate this landscape, both in ways that improve security and in ways that might increase risk. We introduce EVMbench, an evaluation that… ▽ More Smart contracts on public blockchains now manage large amounts of value, and vulnerabilities in these systems can lead to substantial losses. As AI agents become more capable at reading, writing, and running code, it is natural to ask how well they can already navigate this landscape, both in ways that improve security and in ways that might increase risk. We introduce EVMbench, an evaluation that measures the ability of agents to detect, patch, and exploit smart contract vulnerabilities. EVMbench draws on 117 curated vulnerabilities from 40 repositories and, in the most realistic setting, uses programmatic grading based on tests and blockchain state under a local Ethereum execution environment. We evaluate a range of frontier agents and find that they are capable of discovering and exploiting vulnerabilities end-to-end against live blockchain instances. We release code, tasks, and tooling to support continued measurement of these capabilities and future work on security. △ Less

Submitted 5 March, 2026; originally announced March 2026.

QD-PCQA: Quality-Aware Domain Adaptation for Point Cloud Quality Assessment

Authors: Guohua Zhang, Jian Jin, Meiqin Liu, Chao Yao, Weisi Lin

Abstract: No-Reference Point Cloud Quality Assessment (NR-PCQA) still struggles with generalization, primarily due to the scarcity of annotated point cloud datasets. Since the Human Visual System (HVS) drives perceptual quality assessment independently of media types, prior knowledge on quality learned from images can be repurposed for point clouds. This insight motivates adopting Unsupervised Domain Adapta… ▽ More No-Reference Point Cloud Quality Assessment (NR-PCQA) still struggles with generalization, primarily due to the scarcity of annotated point cloud datasets. Since the Human Visual System (HVS) drives perceptual quality assessment independently of media types, prior knowledge on quality learned from images can be repurposed for point clouds. This insight motivates adopting Unsupervised Domain Adaptation (UDA) to transfer quality-relevant priors from labeled images to unlabeled point clouds. However, existing UDA-based PCQA methods often overlook key characteristics of perceptual quality, such as sensitivity to quality ranking and quality-aware feature alignment, thereby limiting their effectiveness. To address these issues, we propose a novel Quality-aware Domain adaptation framework for PCQA, termed QD-PCQA. The framework comprises two main components: i) a Rank-weighted Conditional Alignment (RCA) strategy that aligns features under consistent quality levels and adaptively emphasizes misranked samples to reinforce perceptual quality ranking awareness; and ii) a Quality-guided Feature Augmentation (QFA) strategy, which includes quality-guided style mixup, multi-layer extension, and dual-domain augmentation modules to augment perceptual feature alignment. Extensive cross-domain experiments demonstrate that QD-PCQA significantly improves generalization in NR-PCQA tasks. △ Less

Submitted 16 March, 2026; v1 submitted 3 March, 2026; originally announced March 2026.

Comments: Accepted by CVPR 2026

Proact-VL: A Proactive VideoLLM for Real-Time AI Companions

Authors: Weicai Yan, Yuhong Dai, Qi Ran, Haodong Li, Wang Lin, Hao Liao, Xing Xie, Tao Jin, Jianxun Lian

Abstract: Proactive and real-time interactive experiences are essential for human-like AI companions, yet face three key challenges: (1) achieving low-latency inference under continuous streaming inputs, (2) autonomously deciding when to respond, and (3) controlling both quality and quantity of generated content to meet real-time constraints. In this work, we instantiate AI companions through two gaming sce… ▽ More Proactive and real-time interactive experiences are essential for human-like AI companions, yet face three key challenges: (1) achieving low-latency inference under continuous streaming inputs, (2) autonomously deciding when to respond, and (3) controlling both quality and quantity of generated content to meet real-time constraints. In this work, we instantiate AI companions through two gaming scenarios, commentator and guide, selected for their suitability for automatic evaluation. We introduce the Live Gaming Benchmark, a large-scale dataset with three representative scenarios: solo commentary, co-commentary, and user guidance, and present Proact-VL, a general framework that shapes multimodal language models into proactive, real-time interactive agents capable of human-like environment perception and interaction. Extensive experiments show Proact-VL achieves superior response latency and quality while maintaining strong video understanding capabilities, demonstrating its practicality for real-time interactive applications. △ Less

Submitted 22 March, 2026; v1 submitted 3 March, 2026; originally announced March 2026.

SAE as a Crystal Ball: Interpretable Features Predict Cross-domain Transferability of LLMs without Training

Authors: Qi Zhang, Yifei Wang, Xiaohan Wang, Jiajun Chai, Guojun Yin, Wei Lin, Yisen Wang

Abstract: In recent years, pre-trained large language models have achieved remarkable success across diverse tasks. Besides the pivotal role of self-supervised pre-training, their effectiveness in downstream applications also depends critically on the post-training process, which adapts models to task-specific data and objectives. However, this process inevitably introduces model shifts that can influence p… ▽ More In recent years, pre-trained large language models have achieved remarkable success across diverse tasks. Besides the pivotal role of self-supervised pre-training, their effectiveness in downstream applications also depends critically on the post-training process, which adapts models to task-specific data and objectives. However, this process inevitably introduces model shifts that can influence performance in different domains, and how such shifts transfer remains poorly understood. To open up the black box, we propose the SAE-based Transferability Score (STS), a new metric that leverages sparse autoencoders (SAEs) to forecast post-training transferability. Taking supervised fine-tuning as an example, STS identifies shifted dimensions in SAE representations and calculates their correlations with downstream domains, enabling reliable estimation of transferability \textit{before} fine-tuning. Extensive experiments across multiple models and domains show that STS accurately predicts the transferability of supervised fine-tuning, achieving Pearson correlation coefficients above 0.7 with actual performance changes. Beyond this, we take an initial step toward extending STS to reinforcement learning. We believe that STS can serve as an {\color{black} interpretable} tool for guiding post-training strategies in LLMs. Code is available at https://github.com/PKU-ML/STS. △ Less

Submitted 3 March, 2026; originally announced March 2026.

Length Generalization Bounds for Transformers

Authors: Andy Yang, Pascal Bergsträßer, Georg Zetzsche, David Chiang, Anthony W. Lin

Abstract: Length generalization is a key property of a learning algorithm that enables it to make correct predictions on inputs of any length, given finite training data. To provide such a guarantee, one needs to be able to compute a length generalization bound, beyond which the model is guaranteed to generalize. This paper concerns the open problem of the computability of such generalization bounds for CRA… ▽ More Length generalization is a key property of a learning algorithm that enables it to make correct predictions on inputs of any length, given finite training data. To provide such a guarantee, one needs to be able to compute a length generalization bound, beyond which the model is guaranteed to generalize. This paper concerns the open problem of the computability of such generalization bounds for CRASP, a class of languages which is closely linked to transformers. A positive partial result was recently shown by Chen et al. for CRASP with only one layer and, under some restrictions, also with two layers. We provide complete answers to the above open problem. Our main result is the non-existence of computable length generalization bounds for CRASP (already with two layers) and hence for transformers. To complement this, we provide a computable bound for the positive fragment of CRASP, which we show equivalent to fixed-precision transformers. For both positive CRASP and fixed-precision transformers, we show that the length complexity is exponential, and prove optimality of the bounds. △ Less

Submitted 13 February, 2026; originally announced March 2026.

Surgical Post-Training: Cutting Errors, Keeping Knowledge

Authors: Wenye Lin, Kai Han

Abstract: Enhancing the reasoning capabilities of Large Language Models (LLMs) via post-training is often constrained by the trade-off between efficiency and catastrophic forgetting. While prior research emphasizes the role of on-policy data in mitigating forgetting, we uncover--and validate both theoretically and empirically--an overlooked yet critical mechanism: the implicit regularization inherent in Dir… ▽ More Enhancing the reasoning capabilities of Large Language Models (LLMs) via post-training is often constrained by the trade-off between efficiency and catastrophic forgetting. While prior research emphasizes the role of on-policy data in mitigating forgetting, we uncover--and validate both theoretically and empirically--an overlooked yet critical mechanism: the implicit regularization inherent in Direct Preference Optimization's (DPO) reward estimate. This motivates our Surgical Post-Training (SPoT), a new paradigm designed to optimize reasoning efficiently while preserving learned prior knowledge. SPoT consists of: (1) a data rectification pipeline that employs an Oracle to surgically correct erroneous steps via minimal edits, generating data proximal to the model's distribution; and (2) a reward-based binary cross-entropy objective. Unlike the relative ranking in DPO, this objective treats reasoning correctness as a binary classification problem, enforcing decoupled supervision signals. Empirically, with only 4k rectified math data pairs, SPoT improves Qwen3-8B's accuracy by 6.2% on average across in-domain and OOD tasks, requiring merely 28 minutes of training on 8x H800 GPUs. Code: https://github.com/Visual-AI/SPoT △ Less

Submitted 2 March, 2026; originally announced March 2026.

Comments: 15 pages

Attn-QAT: 4-Bit Attention With Quantization-Aware Training

Authors: Peiyuan Zhang, Matthew Noto, Wenxuan Tan, Chengquan Jiang, Will Lin, Wei Zhou, Hao Zhang

Abstract: Achieving reliable 4-bit attention is a prerequisite for end-to-end FP4 computation on emerging FP4-capable GPUs, yet attention remains the main obstacle due to FP4's tiny dynamic range and attention's heavy-tailed activations. This paper presents the first systematic study of 4-bit quantization-aware training (QAT) for attention. We find that "drop-in" QAT, which naively combines an FP4 forward p… ▽ More Achieving reliable 4-bit attention is a prerequisite for end-to-end FP4 computation on emerging FP4-capable GPUs, yet attention remains the main obstacle due to FP4's tiny dynamic range and attention's heavy-tailed activations. This paper presents the first systematic study of 4-bit quantization-aware training (QAT) for attention. We find that "drop-in" QAT, which naively combines an FP4 forward pass with a high-precision Flash Attention (FA)-style backward pass, leads to training instability. We identify two key principles for stable FP4 attention: (1) matching low-precision recomputation of attention scores in the backward pass, and (2) resolving implicit precision assumptions in FA's gradient calculation. Based on these insights, we propose Attn-QAT and implement fused Triton kernels for training as well as FP4 inference kernels. Across diffusion and language models, Attn-QAT recovers the quality drop from FP4 attention without explicit outlier-mitigation heuristics used in prior FP4 attention, and delivers up to a 1.5x speedup on an RTX 5090. Video demos can be found at https://drive.google.com/drive/folders/190F6xbBDUF2kGQYIcXBt3ehSYij5jlim?usp=sharing. △ Less

Submitted 6 March, 2026; v1 submitted 8 February, 2026; originally announced March 2026.

Higress-RAG: A Holistic Optimization Framework for Enterprise Retrieval-Augmented Generation via Dual Hybrid Retrieval, Adaptive Routing, and CRAG

Authors: Weixi Lin

Abstract: The integration of Large Language Models (LLMs) into enterprise knowledge management systems has been catalyzed by the Retrieval-Augmented Generation (RAG) paradigm, which augments parametric memory with non-parametric external data. However, the transition from proof-of-concept to production-grade RAG systems is hindered by three persistent challenges: low retrieval precision for complex queries,… ▽ More The integration of Large Language Models (LLMs) into enterprise knowledge management systems has been catalyzed by the Retrieval-Augmented Generation (RAG) paradigm, which augments parametric memory with non-parametric external data. However, the transition from proof-of-concept to production-grade RAG systems is hindered by three persistent challenges: low retrieval precision for complex queries, high rates of hallucination in the generation phase, and unacceptable latency for real-time applications. This paper presents a comprehensive analysis of the Higress RAG MCP Server, a novel, enterprise-centric architecture designed to resolve these bottlenecks through a "Full-Link Optimization" strategy. Built upon the Model Context Protocol (MCP), the system introduces a layered architecture that orchestrates a sophisticated pipeline of Adaptive Routing, Semantic Caching, Hybrid Retrieval, and Corrective RAG (CRAG). We detail the technical implementation of key innovations, including the Higress-Native Splitter for structure-aware data ingestion, the application of Reciprocal Rank Fusion (RRF) for merging dense and sparse retrieval signals, and a 50ms-latency Semantic Caching mechanism with dynamic thresholding. Experimental evaluations on domain-specific Higress technical documentation and blogs verify the system's architectural robustness. The results demonstrate that by optimizing the entire retrieval lifecycle - from pre-retrieval query rewriting to post-retrieval corrective evaluation - the Higress RAG system offers a scalable, hallucination-resistant solution for enterprise AI deployment. △ Less

Submitted 30 December, 2025; originally announced February 2026.

Comments: 7 pages,5 figures, our submissions are not yet published

ACM Class: H.3.3; I.2.7; D.2.11

Efficient Encoder-Free Fourier-based 3D Large Multimodal Model

Authors: Guofeng Mei, Wei Lin, Luigi Riz, Yujiao Wu, Yiming Wang, Fabio Poiesi

Abstract: Large Multimodal Models (LMMs) that process 3D data typically rely on heavy, pre-trained visual encoders to extract geometric features. While recent 2D LMMs have begun to eliminate such encoders for efficiency and scalability, extending this paradigm to 3D remains challenging due to the unordered and large-scale nature of point clouds. This leaves a critical unanswered question: How can we design… ▽ More Large Multimodal Models (LMMs) that process 3D data typically rely on heavy, pre-trained visual encoders to extract geometric features. While recent 2D LMMs have begun to eliminate such encoders for efficiency and scalability, extending this paradigm to 3D remains challenging due to the unordered and large-scale nature of point clouds. This leaves a critical unanswered question: How can we design an LMM that tokenizes unordered 3D data effectively and efficiently without a cumbersome encoder? We propose Fase3D, the first efficient encoder-free Fourier-based 3D scene LMM. Fase3D tackles the challenges of scalability and permutation invariance with a novel tokenizer that combines point cloud serialization and the Fast Fourier Transform (FFT) to approximate self-attention. This design enables an effective and computationally minimal architecture, built upon three key innovations: First, we represent large scenes compactly via structured superpoints. Second, our space-filling curve serialization followed by an FFT enables efficient global context modeling and graph-based token merging. Lastly, our Fourier-augmented LoRA adapters inject global frequency-aware interactions into the LLMs at a negligible cost. Fase3D achieves performance comparable to encoder-based 3D LMMs while being significantly more efficient in computation and parameters. Project website: https://tev-fbk.github.io/Fase3D. △ Less

Submitted 28 March, 2026; v1 submitted 26 February, 2026; originally announced February 2026.

Journal ref: CVPR 2026 camera ready

Scaling Audio-Visual Quality Assessment Dataset via Crowdsourcing

Authors: Renyu Yang, Jian Jin, Lili Meng, Meiqin Liu, Yilin Wang, Balu Adsumilli, Weisi Lin

Abstract: Audio-visual quality assessment (AVQA) research has been stalled by limitations of existing datasets: they are typically small in scale, with insufficient diversity in content and quality, and annotated only with overall scores. These shortcomings provide limited support for model development and multimodal perception research. We propose a practical approach for AVQA dataset construction. First,… ▽ More Audio-visual quality assessment (AVQA) research has been stalled by limitations of existing datasets: they are typically small in scale, with insufficient diversity in content and quality, and annotated only with overall scores. These shortcomings provide limited support for model development and multimodal perception research. We propose a practical approach for AVQA dataset construction. First, we design a crowdsourced subjective experiment framework for AVQA, breaks the constraints of in-lab settings and achieves reliable annotation across varied environments. Second, a systematic data preparation strategy is further employed to ensure broad coverage of both quality levels and semantic scenarios. Third, we extend the dataset with additional annotations, enabling research on multimodal perception mechanisms and their relation to content. Finally, we validate this approach through YT-NTU-AVQ, the largest and most diverse AVQA dataset to date, consisting of 1,620 user-generated audio and video (A/V) sequences. The dataset and platform code are available at https://github.com/renyu12/YT-NTU-AVQ △ Less

Submitted 26 February, 2026; originally announced February 2026.

Comments: Accepted to ICASSP 2026. 5 pages (main paper) + 8 pages (supplementary material)

BoxSplitGen: A Generative Model for 3D Part Bounding Boxes in Varying Granularity

Authors: Juil Koo, Wei-Tung Lin, Chanho Park, Chanhyeok Park, Minhyuk Sung

Abstract: Human creativity follows a perceptual process, moving from abstract ideas to finer details during creation. While 3D generative models have advanced dramatically, models specifically designed to assist human imagination in 3D creation -- particularly for detailing abstractions from coarse to fine -- have not been explored. We propose a framework that enables intuitive and interactive 3D shape gene… ▽ More Human creativity follows a perceptual process, moving from abstract ideas to finer details during creation. While 3D generative models have advanced dramatically, models specifically designed to assist human imagination in 3D creation -- particularly for detailing abstractions from coarse to fine -- have not been explored. We propose a framework that enables intuitive and interactive 3D shape generation by iteratively splitting bounding boxes to refine the set of bounding boxes. The main technical components of our framework are two generative models: the box-splitting generative model and the box-to-shape generative model. The first model, named BoxSplitGen, generates a collection of 3D part bounding boxes with varying granularity by iteratively splitting coarse bounding boxes. It utilizes part bounding boxes created through agglomerative merging and learns the reverse of the merging process -- the splitting sequences. The model consists of two main components: the first learns the categorical distribution of the box to be split, and the second learns the distribution of the two new boxes, given the set of boxes and the indication of which box to split. The second model, the box-to-shape generative model, is trained by leveraging the 3D shape priors learned by an existing 3D diffusion model while adapting the model to incorporate bounding box conditioning. In our experiments, we demonstrate that the box-splitting generative model outperforms token prediction models and the inpainting approach with an unconditional diffusion model. Also, we show that our box-to-shape model, based on a state-of-the-art 3D diffusion model, provides superior results compared to a previous model. △ Less

Submitted 24 February, 2026; originally announced February 2026.

Comments: Project page: https://boxsplitgen.github.io

Using Unsupervised Domain Adaptation Semantic Segmentation for Pulmonary Embolism Detection in Computed Tomography Pulmonary Angiogram (CTPA) Images

Authors: Wen-Liang Lin, Yun-Chien Cheng

Abstract: While deep learning has demonstrated considerable promise in computer-aided diagnosis for pulmonary embolism (PE), practical deployment in Computed Tomography Pulmonary Angiography (CTPA) is often hindered by "domain shift" and the prohibitive cost of expert annotations. To address these challenges, an unsupervised domain adaptation (UDA) framework is proposed, utilizing a Transformer backbone and… ▽ More While deep learning has demonstrated considerable promise in computer-aided diagnosis for pulmonary embolism (PE), practical deployment in Computed Tomography Pulmonary Angiography (CTPA) is often hindered by "domain shift" and the prohibitive cost of expert annotations. To address these challenges, an unsupervised domain adaptation (UDA) framework is proposed, utilizing a Transformer backbone and a Mean-Teacher architecture for cross-center semantic segmentation. The primary focus is placed on enhancing pseudo-label reliability by learning deep structural information within the feature space. Specifically, three modules are integrated and designed for this task: (1) a Prototype Alignment (PA) mechanism to reduce category-level distribution discrepancies; (2) Global and Local Contrastive Learning (GLCL) to capture both pixel-level topological relationships and global semantic representations; and (3) an Attention-based Auxiliary Local Prediction (AALP) module designed to reinforce sensitivity to small PE lesions by automatically extracting high-information slices from Transformer attention maps. Experimental validation conducted on cross-center datasets (FUMPE and CAD-PE) demonstrates significant performance gains. In the FUMPE -> CAD-PE task, the IoU increased from 0.1152 to 0.4153, while the CAD-PE -> FUMPE task saw an improvement from 0.1705 to 0.4302. Furthermore, the proposed method achieved a 69.9% Dice score in the CT -> MRI cross-modality task on the MMWHS dataset without utilizing any target-domain labels for model selection, confirming its robustness and generalizability for diverse clinical environments. △ Less

Submitted 23 February, 2026; originally announced February 2026.

Synthesis and Verification of Transformer Programs

Authors: Hongjian Jiang, Matthew Hague, Philipp Rümmer, Anthony Widjaja Lin

Abstract: C-RASP is a simple programming language that was recently shown to capture concepts expressible by transformers. In this paper, we develop new algorithmic techniques for automatically verifying C-RASPs. To this end, we establish a connection to the verification of synchronous dataflow programs in Lustre, which enables us to exploit state-of-the-art model checkers utilizing highly optimized SMT-sol… ▽ More C-RASP is a simple programming language that was recently shown to capture concepts expressible by transformers. In this paper, we develop new algorithmic techniques for automatically verifying C-RASPs. To this end, we establish a connection to the verification of synchronous dataflow programs in Lustre, which enables us to exploit state-of-the-art model checkers utilizing highly optimized SMT-solvers. Our second contribution addresses learning a C-RASP program in the first place. To this end, we provide a new algorithm for learning a C-RASP from examples using local search. We demonstrate efficacy of our implementation for benchmarks of C-RASPs in the literature, in particular in connection to the following applications: (1) transformer program optimization, and (2) constrained learning of transformer programs (based on a partial specification). △ Less

Submitted 18 February, 2026; originally announced February 2026.

Explainable Token-level Noise Filtering for LLM Fine-tuning Datasets

Authors: Yuchen Yang, Wenze Lin, Enhao Huang, Zhixuan Chu, Hongbin Zhou, Lan Tao, Yiming Li, Zhan Qin, Kui Ren

Abstract: Large Language Models (LLMs) have seen remarkable advancements, achieving state-of-the-art results in diverse applications. Fine-tuning, an important step for adapting LLMs to specific downstream tasks, typically involves further training on corresponding datasets. However, a fundamental discrepancy exists between current fine-tuning datasets and the token-level optimization mechanism of LLMs: mos… ▽ More Large Language Models (LLMs) have seen remarkable advancements, achieving state-of-the-art results in diverse applications. Fine-tuning, an important step for adapting LLMs to specific downstream tasks, typically involves further training on corresponding datasets. However, a fundamental discrepancy exists between current fine-tuning datasets and the token-level optimization mechanism of LLMs: most datasets are designed at the sentence-level, which introduces token-level noise, causing negative influence to final performance. In this paper, we propose XTF, an explainable token-level noise filtering framework. XTF decomposes the complex and subtle contributions of token-level data to the fine-tuning process into three distinct and explicit attributes (reasoning importance, knowledge novelty, and task relevance), which can be assessed using scoring methods, and then masks the gradients of selected noisy tokens accordingly to optimize the performance of fine-tuned LLMs. We conduct extensive experiments on three representative downstream tasks (math, code and medicine) across 7 mainstream LLMs. The results demonstrate that XTF can significantly improve downstream performance by up to 13.7% compared to regular fine-tuning. Our work highlights the importance of token-level dataset optimization, and demonstrates the potential of strategies based on attribute decomposition for explaining complex training mechanisms. △ Less

Submitted 5 April, 2026; v1 submitted 16 February, 2026; originally announced February 2026.