❌

Normal view

Received — 2 April 2026 ⏭

MindCube: Spatial Mental Modeling from Limited Views

✇cs.AI, q-bio.NC updates on arXiv.org
By: Qineng Wang · Baiqiao Yin · Pingyue Zhang · Jianshu Zhang · Kangrui Wang · Zihan Wang · Jieyu Zhang · Keshigeyan Chandrasegaran · Han Liu · Ranjay Krishna · Saining Xie · Jiajun Wu · Li Fei-Fei · Manling Li
1 April 2026 at 12:00
arXiv:2506.21458v2 Announce Type: replace Abstract: Can Vision-Language Models (VLMs) imagine the full scene from just a few views, like humans do? Humans form spatial mental models naturally, internal representations of unseen space, to reason about layout, perspective, and motion. Our MindCube benchmark with 21,154 questions across 3,268 images exposes this critical gap, where existing VLMs exhibit near-random performance. Using MindCube, we systematically evaluate how well VLMs build robust spatial mental models through representing positions (cognitive mapping), orientations (perspective-taking), and dynamics (mental simulation for "what-if" movements). We then explore three approaches to help approximate spatial mental models in VLMs, focusing on incorporating unseen intermediate views, natural language reasoning chains, and cognitive maps. The significant improvement comes from a synergistic approach, "map-then-reason", that jointly trains the model to first generate a cognitive map and then reason upon it. By training models to reason over these internal maps, we boosted accuracy from 37.8% to 57.8% (+20.0%). Adding reinforcement learning pushed performance even further to 61.3% (+23.5%). Our key insight is that such scaffolding of spatial mental models, actively constructing and utilizing internal structured spatial representations with flexible reasoning processes, significantly improves understanding of unobservable space.

InCoder-32B: Code Foundation Model for Industrial Scenarios

✇cs.AI, q-bio.NC updates on arXiv.org
By: Jian Yang · Wei Zhang · Jiajun Wu · Junhang Cheng · Shawn Guo · Haowen Wang · Weicheng Gu · Yaxin Du · Joseph Li · Fanglin Xu · Yizhi Li · Lin Jing · Yuanbo Wang · Yuhan Gao · Ruihao Gong · Chuan Hao · Ran Tao · Aishan Liu · Tuney Zheng · Ganqu Cui · Zhoujun Li · Mingjie Tang · Chenghua Lin · Wayne Xin Zhao · Xianglong Liu · Ming Zhou · Bryan Dai · Weifeng Lv
1 April 2026 at 12:00
arXiv:2603.16790v3 Announce Type: replace-cross Abstract: Recent code large language models have achieved remarkable progress on general programming tasks. Nevertheless, their performance degrades significantly in industrial scenarios that require reasoning about hardware semantics, specialized language constructs, and strict resource constraints. To address these challenges, we introduce InCoder-32B (Industrial-Coder-32B), the first 32B-parameter code foundation model unifying code intelligence across chip design, GPU kernel optimization, embedded systems, compiler optimization, and 3D modeling. By adopting an efficient architecture, we train InCoder-32B from scratch with general code pre-training, curated industrial code annealing, mid-training that progressively extends context from 8K to 128K tokens with synthetic industrial reasoning data, and post-training with execution-grounded verification. We conduct extensive evaluation on 14 mainstream general code benchmarks and 9 industrial benchmarks spanning 4 specialized domains. Results show InCoder-32B achieves highly competitive performance on general tasks while establishing strong open-source baselines across industrial domains.

Gene regulatory landscape dissected by single-cell four-omics sequencing

✇Nature - Issue - nature.com science feeds
By: Yujie Chen · Zhiyuan Liu · Heming Xu · Jiayu Liu · Mengxuan Wang · Yi Chi · Boyuan Liang · Menghan Liu · Yongli Peng · Hao Ge · Dong Xing
1 April 2026 at 08:00

Nature, Published online: 01 April 2026; doi:10.1038/s41586-026-10322-z

Combining single-cell parallel profiling of genome conformation, histone modifications, chromatin accessibility and gene expression reveals dynamics and intranuclear spatial clustering of epigenome profiles, enabling sophisticated analysis of the regulatory landscape across cell types and tissues.

Nanoscale transfer-printed full-colour ultrahigh-resolution quantum dot LEDs

✇Nature - Issue - nature.com science feeds
By: Lihua Lin · Jie Wang · Hailong Hu · Haolin Luo · Yanbin Liu · Xingjie Yang · Jingnan Su · De’er Li · Zhongwei Xu · Chengyu Luo · Yongshen Yu · Tailiang Guo · Fushan Li
1 April 2026 at 08:00

Nature, Published online: 01 April 2026; doi:10.1038/s41586-026-10333-w

A dual-action force dynamics strategy using a hard silicon template as a nanoimprinting stamp combined with inverted transfer printing is described for the manufacture of high-performance full-colour ultrahigh-resolution quantum dot light-emitting diodes (LEDs) for active-matrix displays, while revealing electric-field reconstruction in nanoscale arrays and introducing dielectric matching to mitigate field concentration and performance degradation.
Received — 26 March 2026 ⏭

From Context to Intent: Reasoning-Guided Function-Level Code Completion

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yanzhou Li · Tianlin Li · Yiran Zhang · Shangqing Liu · Aishan Liu · Xianglong Liu · Yang Liu
25 March 2026 at 12:00
arXiv:2508.09537v2 Announce Type: replace-cross Abstract: The growing capabilities of Large Language Models (LLMs) have led to their widespread adoption for function completion within code repositories. Recent studies on such tasks show promising results when explicit instructions, often in the form of docstrings, are available to guide the completion. However, in real-world scenarios, clear docstrings are frequently absent. Under such conditions, LLMs typically fail to produce accurate completions. To enable more automated and accurate function completion in such settings, we aim to enable LLMs to accurately infer the developer's intent prior to code completion. Our key insight is that the preceding code, namely the code context before the function to be completed, often contains valuable cues that help the model understand the intended functionality. However, inferring intent from such implicit context is non-trivial and constitutes a core challenge in function-level code completion. To tackle this challenge, inspired by how humans interpret context, we propose a reasoning-based prompting framework that guides LLMs to utilize these contextual cues to infer intent step by step. To incentivize LLMs to reason through the preceding code and infer intent, we further curate a dataset of 40k examples, each annotated with intermediate reasoning traces and corresponding docstrings. Extensive experiments on DevEval and ComplexCodeEval demonstrate consistent performance improvements across multiple models, achieving over 25% relative gains in pass@1 for both DeepSeekCoder and CodeLLaMA families. Building upon our framework, we further develop an intent-interactive platform that supports lightweight human feedback. This platform allows developers to select from a set of candidate intentions or edit the intent to better guide the model. Our experiments show that this interactive approach leads to further performance improvements.

The Current Landscape of Remote Digital Symptom Monitoring for Patients With Lung Cancer: Scoping Review

✇Journal of Medical Internet Research
By: Hao Huang · Naomi Takemura · Hammoda Abu-Odah · Shuhan Li · Ting Mao · Zengjie Ye · Janelle Yorke
25 March 2026 at 02:00
Background: Remote digital symptom monitoring systems (rSMS) have been increasingly used in recent years to monitor symptoms, health-related quality of life, and other patient-reported outcomes in lung cancer. Previous studies have demonstrated variability in study design, types of rSMS, and outcomes used to assess benefits for patients and health care systems. However, there remains a lack of synthesized evidence pertaining to the similarities and differences among rSMS, including their theoretical underpinnings, key functional components, and reported benefits and limitations. Objective: This review aims to identify and synthesize existing research to map the current landscape of rSMS in lung cancer, including the theoretical foundations for its development and implementation, as well as its types, applications, and outcomes. Methods: This scoping review followed the Joanna Briggs Institute scoping review framework and adhered to the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) guidelines. A comprehensive literature search was conducted from database inception to October 16, 2025, across 7 English-language databases and 3 Chinese-language databases (CNKI, WanFang, and SinoMed). Eligible studies were peer-reviewed original research articles examining rSMS among adults with lung cancer. Data were independently screened and extracted by 2 reviewers, with discrepancies resolved by a third reviewer. Quantitative data were extracted using a standardized form and synthesized descriptively. Content analysis was performed to analyze the qualitative data. Results: A total of 41 studies involving 11,765 patients and 85 health care providers were included. Twelve studies focused exclusively on advanced-stage lung cancer. Participants were generally middle-aged to older adults (mean ages 51‐74 y), with male participants typically comprising 30% to 50% across studies. Most studies were conducted in the United States (n=19). We identified 32 patient-reported outcome measures that were used either as core rSMS components or as study outcomes. Four common functional modules were observed across rSMS: data collection, data analysis, response systems, and patient education. Qualitative evidence was limited; the most frequently reported benefit was the promotion of patient-centered care. Health care providers raised concerns about uncertain effectiveness and increased workload. Conclusions: This scoping review highlights the promising role of rSMS in lung cancer care and provides a structured map of current evidence. It adds to prior literature in 3 ways. First, it summarizes how and how often theoretical frameworks are reported and applied in rSMS development and implementation. Second, it synthesizes and categorizes four common functional modules across systems. Third, it differentiates measures embedded as rSMS components from those used as evaluation outcomes. These contributions clarify current practices and methodological gaps and underscore the importance of theory-informed design, functional clarity, and stakeholder engagement in the development of patient-centered, clinically meaningful, and sustainable rSMS platforms. Trial Registration: OSF Registries 9637t; https://osf.io/9637t/overview

β-hydroxybutyrate enhances the metabolic fitness of CAR T cells in cancer

✇Cell
By: Shan Liu · Puneeth Guruprasad · Ranjani Ramasubramanian · Bhoomi Madhu · Luca Paruzzo · Kecheng Han · Andre Kelly · Alexander Shestov · He N. Xu · Alberto Carturan · Chaoting Zhou · Kevin R. Amses · Amichay Afriat · Lev Litichevskiy · Jason Lin · Ezra Dubowitz · Neil Tangal · Jing Zhang · Alana McSween · Melody Tan · Federico Stella · Andrew Lee · Siena Nason · Xianxin Hua · Michael Schneider · Madeleine Sleeman · Yunlin Zhang · Giulia Gabrielli · Ziqi Yang · Raymone Pajarillo · Ruchi Patel · Guido Ghilardi · Vrutti Patel · Akshita Joshi · Shunzhou Jiang · Yanqing Jiang · Patrizia Porazzi · Julia C. Tchou · Brian Keith · Mingyao Li · Elise Chong · Stephen J. Schuster · Michael Milone · Joshua Rabinowitz · Roddy S. O’Connor · Christoph A. Thaiss · Maayan Levy · Marco Ruella
6 March 2026 at 08:00
β-hydroxybutyrate (BHB), the ketone body associated with a ketogenic diet, metabolically reprograms and fuels CAR T cells to achieve proliferation, cytokine production, and superior tumor control. These findings suggest that BHB supplementation may be a practical way to boost adoptive cancer immunotherapy.
Received — 17 March 2026 ⏭

VideoTemp-o3: Harmonizing Temporal Grounding and Video Understanding in Agentic Thinking-with-Videos

✇cs.AI, q-bio.NC updates on arXiv.org
By: Wenqi Liu · Yunxiao Wang · Shijie Ma · Meng Liu · Qile Su · Tianke Zhang · Haonan Fan · Changyi Liu · Kaiyu Jiang · Jiankang Chen · Kaiyu Tang · Bin Wen · Fan Yang · Tingting Gao · Han Li · Yinwei Wei · Xuemeng Song
16 March 2026 at 12:00
arXiv:2602.07801v3 Announce Type: replace-cross Abstract: In long-video understanding, conventional uniform frame sampling often fails to capture key visual evidence, leading to degraded performance and increased hallucinations. To address this, recent agentic thinking-with-videos paradigms have emerged, adopting a localize-clip-answer pipeline in which the model actively identifies relevant video segments, performs dense sampling within those clips, and then produces answers. However, existing methods remain inefficient, suffer from weak localization, and adhere to rigid workflows. To solve these issues, we propose VideoTemp-o3, a unified agentic thinking-with-videos framework that jointly models video grounding and question answering. VideoTemp-o3 exhibits strong localization capability, supports on-demand clipping, and can refine inaccurate localizations. Specifically, in the supervised fine-tuning stage, we design a unified masking mechanism that encourages exploration while preventing noise. For reinforcement learning, we introduce dedicated rewards to mitigate reward hacking. Besides, from the data perspective, we develop an effective pipeline to construct high-quality long video grounded QA data, along with a corresponding benchmark for systematic evaluation across various video durations. Experimental results demonstrate that our method achieves remarkable performance on both long video understanding and grounding.

Automatic In-Domain Exemplar Construction and LLM-Based Refinement of Multi-LLM Expansions for Query Expansion

✇cs.AI, q-bio.NC updates on arXiv.org
By: Minghan Li · Ercong Nie · Siqi Zhao · Tongna Chen · Huiping Huang · Guodong Zhou
16 March 2026 at 12:00
arXiv:2602.08917v2 Announce Type: replace-cross Abstract: Query expansion with large language models is promising but often relies on hand-crafted prompts, manually chosen exemplars, or a single LLM, making it non-scalable and sensitive to domain shift. We present an automated, domain-adaptive QE framework that builds in-domain exemplar pools by harvesting pseudo-relevant passages using a BM25-MonoT5 pipeline. A training-free cluster-based strategy selects diverse demonstrations, yielding strong and stable in-context QE without supervision. To further exploit model complementarity, we introduce a two-LLM ensemble in which two heterogeneous LLMs independently generate expansions and a refinement LLM consolidates them into one coherent expansion. Across TREC DL20, DBPedia, and SciFact, the refined ensemble delivers consistent and statistically significant gains over BM25, Rocchio, zero-shot, and fixed few-shot baselines. The framework offers a reproducible testbed for exemplar selection and multi-LLM generation, and a practical, label-free solution for real-world QE.
Received — 14 March 2026 ⏭

Functional-based multi-omics early prediction of radiation pneumonitis in NSCLC using AI-generated perfusion and ventilation from planning CT

✇(Multiomics OR Omics) AND (Lung OR gastric OR Hepatocellular)
By: Mayang Zhao · Tao Peng · Zhi Chen · Tianyu Xiong · Bing Li · Xiaoli Zheng · Yu-Hua Huang · Liming Song · Yao Pu · Zihan Li · Jing Cai · Ge Ren
13 March 2026 at 18:00

Phys Med Biol. 2026 Mar 13. doi: 10.1088/1361-6560/ae5209. Online ahead of print.

ABSTRACT

ObjectiveThis study aims to develop a functional-based multi-omics model for early prediction of radiation pneumonitis (RP) by extracting radiomic and dosiomic features from functionally defined lung regions, using generated perfusion (Q) and ventilation (V) from pre-radiotherapy planning computed tomography (CT).
ApproachWe retrospectively analyzed data from 121 patients with locally advanced non-small cell lung cancer (NSCLC) treated with curative-intent IMRT between 2015 and 2019, including pre-treatment CT and dose maps. Q and V maps were generated from CT with deep learning-based and supervoxel-based approaches, respectively. Regions of interest (ROIs) combined the planning target volume (PTV) with each of three functional lung regions-high functional lung (HFL), low functional lung (LFL), and whole lung (WL)-defined by thresholds on Q and V maps. Radiomic and dosiomic features were extracted from CT and dose distributions within each ROI. For each ROI, For each ROI, three methods-radiomics (R), dosiomics (D), and dual-omics (RD)-were constructed. 13 machine learning algorithms were trained and evaluated using 10-fold cross-validation, and model performance was assessed by the average area under the receiver operating characteristic curve (AUC), accuracy, precision, recall, and F1 score. RP was defined as CTCAE grade ≥ 2.
Main resultsOf the 35 selected features, 20 were from HFL. In dual-omics models, using HFL features improved predictive performance for RP (AUC 0.879±0.105) compared to WL (AUC 0.778 ± 0.100). In HFL, the RD method outperformed both R (AUC 0.786± 0.076) and D (AUC 0.791 ± 0.107) methods. Decision curve analysis showed the dual-omics model based on HFL provided the highest net benefit across threshold probabilities.
SignificanceThis study is the first to systematically demonstrate that features extracted from CT-derived HFL capture important functional differences and provide strong predictive value for RP. Compared to conventional methods, integrating radiomics, dosiomics, and CT-based functional information further improves predictive performance.&#xD.

PMID:41825133 | DOI:10.1088/1361-6560/ae5209

Targeting the OXNAD1-PTGS2 axis with resveratrol overcomes ferroptosis Inhibition and reverses 5-FU resistance in gastric cancer

✇Omics in Gastric
By: Jinzhou Wang · Bishuang Xu · Xiaohan Lin · Biyu Chen · Xiaoyan Liu · Wei Wu
13 March 2026 at 18:00

Gastric Cancer. 2026 Mar 13. doi: 10.1007/s10120-026-01718-x. Online ahead of print.

ABSTRACT

BACKGROUND: 5-Fluorouracil (5-FU) remains a cornerstone of first-line chemotherapy for gastric cancer, yet the emergence of resistance severely compromises its clinical efficacy. Although ferroptosis suppression has been recognized as a pivotal mechanism of chemoresistance, the mitochondrial regulatory processes involved remain poorly understood.

METHODS: We integrated clinical specimen analysis, in vitro and in vivo functional assays, multi-omics profiling, and molecular docking to delineate the role of the mitochondrial oxidoreductase OXNAD1 in mediating 5-FU resistance in gastric cancer, and to assess the therapeutic potential of the natural polyphenol resveratrol as a chemosensitizing agent.

RESULTS: OXNAD1 was found to be significantly overexpressed in gastric cancer tissues and cell lines, correlating with unfavorable prognosis and enhanced 5-FU resistance. Mechanistically, OXNAD1 directly bound to and suppressed the ferroptosis driver PTGS2, thereby attenuating lipid peroxidation and mitochondrial damage, ultimately restraining ferroptosis and promoting drug resistance. Notably, resveratrol disrupted the OXNAD1-PTGS2 interaction by directly binding OXNAD1, reinstating ferroptotic activity, markedly enhancing the cytotoxic effect of 5-FU in resistant cells, and potentiating the antitumor efficacy of 5-FU in xenograft models.

CONCLUSION: The OXNAD1-PTGS2 axis constitutes a critical metabolic-cell death cross-regulatory pathway underlying 5-FU resistance in gastric cancer. Targeting this axis with resveratrol provides a promising combinatorial strategy to overcome chemoresistance.

PMID:41824193 | DOI:10.1007/s10120-026-01718-x

Functional-based multi-omics early prediction of radiation pneumonitis in NSCLC using AI-generated perfusion and ventilation from planning CT

✇Omics In Lung
By: Mayang Zhao · Tao Peng · Zhi Chen · Tianyu Xiong · Bing Li · Xiaoli Zheng · Yu-Hua Huang · Liming Song · Yao Pu · Zihan Li · Jing Cai · Ge Ren
13 March 2026 at 18:00

Phys Med Biol. 2026 Mar 13. doi: 10.1088/1361-6560/ae5209. Online ahead of print.

ABSTRACT

ObjectiveThis study aims to develop a functional-based multi-omics model for early prediction of radiation pneumonitis (RP) by extracting radiomic and dosiomic features from functionally defined lung regions, using generated perfusion (Q) and ventilation (V) from pre-radiotherapy planning computed tomography (CT).
ApproachWe retrospectively analyzed data from 121 patients with locally advanced non-small cell lung cancer (NSCLC) treated with curative-intent IMRT between 2015 and 2019, including pre-treatment CT and dose maps. Q and V maps were generated from CT with deep learning-based and supervoxel-based approaches, respectively. Regions of interest (ROIs) combined the planning target volume (PTV) with each of three functional lung regions-high functional lung (HFL), low functional lung (LFL), and whole lung (WL)-defined by thresholds on Q and V maps. Radiomic and dosiomic features were extracted from CT and dose distributions within each ROI. For each ROI, For each ROI, three methods-radiomics (R), dosiomics (D), and dual-omics (RD)-were constructed. 13 machine learning algorithms were trained and evaluated using 10-fold cross-validation, and model performance was assessed by the average area under the receiver operating characteristic curve (AUC), accuracy, precision, recall, and F1 score. RP was defined as CTCAE grade ≥ 2.
Main resultsOf the 35 selected features, 20 were from HFL. In dual-omics models, using HFL features improved predictive performance for RP (AUC 0.879±0.105) compared to WL (AUC 0.778 ± 0.100). In HFL, the RD method outperformed both R (AUC 0.786± 0.076) and D (AUC 0.791 ± 0.107) methods. Decision curve analysis showed the dual-omics model based on HFL provided the highest net benefit across threshold probabilities.
SignificanceThis study is the first to systematically demonstrate that features extracted from CT-derived HFL capture important functional differences and provide strong predictive value for RP. Compared to conventional methods, integrating radiomics, dosiomics, and CT-based functional information further improves predictive performance.&#xD.

PMID:41825133 | DOI:10.1088/1361-6560/ae5209

Received — 11 March 2026 ⏭

Grouter: Decoupling Routing from Representation for Accelerated MoE Training

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yuqi Xu · Rizhen Hu · Zihan Liu · Mou Sun · Kun Yuan
10 March 2026 at 12:00
arXiv:2603.06626v1 Announce Type: cross Abstract: Traditional Mixture-of-Experts (MoE) training typically proceeds without any structural priors, effectively requiring the model to simultaneously train expert weights while searching for an optimal routing policy within a vast combinatorial space. This entanglement often leads to sluggish convergence and training instabilities. This paper introduces Grouter, a preemptive routing method that by distilling high-quality structures from fully-trained MoE models and serving as a fixed router for target models. By decoupling structural optimization from weight updates, Grouter significantly accelerates both the speed and quality of model convergence. To ensure the framework's versatility, we also introduce expert folding to adapt Grouter across varying model configurations and expert tuning to rebalance workloads across different data distributions. Furthermore, by leveraging the structural priors provided by preemptive routing, we can implement targeted optimizations to further enhance training throughput. Experiments demonstrate that Grouter achieves superior performance and efficiency which boosts pre-training data utilization by 4.28x and achieves up to 33.5% throughput acceleration, establishing preemptive routing as a fundamental paradigm for scalable MoE training.

Hide and Find: A Distributed Adversarial Attack on Federated Graph Learning

✇cs.AI, q-bio.NC updates on arXiv.org
By: Jinshan Liu · Ken Li · Jiazhe Wei · Bin Shi · Bo Dong
10 March 2026 at 12:00
arXiv:2603.07743v1 Announce Type: cross Abstract: Federated Graph Learning (FedGL) is vulnerable to malicious attacks, yet developing a truly effective and stealthy attack method remains a significant challenge. Existing attack methods suffer from low attack success rates, high computational costs, and are easily identified and smoothed by defense algorithms. To address these challenges, we propose \textbf{FedShift}, a novel two-stage "Hide and Find" distributed adversarial attack. In the first stage, before FedGL begins, we inject a learnable and hidden "shifter" into part of the training data, which subtly pushes poisoned graph representations toward a target class's decision boundary without crossing it, ensuring attack stealthiness during training. In the second stage, after FedGL is complete, we leverage the global model information and use the hidden shifter as an optimization starting point to efficiently find the adversarial perturbations. During the final attack, we aggregate these perturbations from multiple malicious clients to form the final effective adversarial sample and trigger the attack. Extensive experiments on six large-scale datasets demonstrate that our method achieves the highest attack effectiveness compared to existing advanced attack methods. In particular, our attack can effectively evade 3 mainstream robust federated learning defense algorithms and converges with a time cost reduction of over 90\%, highlighting its exceptional stealthiness, robustness, and efficiency.

Dexamethasone promotes neutrophil ROS-mediated tumor killing through the glucocorticoid receptor

✇Oncogene - Issue - nature.com science feeds
By: Zhanhong Liu · Rongrong Sun · Yinghong Li · Ziqi Zhang · Peiqing Huang · Yipeng Zhou · Pengbo Hou · Wenqing Mu · Gerry Melino · Peishan Li · Yufang Shi · Changshun Shao
6 March 2026 at 08:00

Oncogene, Published online: 06 March 2026; doi:10.1038/s41388-026-03708-w

Dexamethasone promotes neutrophil ROS-mediated tumor killing through the glucocorticoid receptor
Received — 5 March 2026 ⏭

Phys4D: Fine-Grained Physics-Consistent 4D Modeling from Video Diffusion

✇cs.AI, q-bio.NC updates on arXiv.org
By: Haoran Lu · Shang Wu · Jianshu Zhang · Maojiang Su · Guo Ye · Chenwei Xu · Lie Lu · Pranav Maneriker · Fan Du · Manling Li · Zhaoran Wang · Han Liu
5 March 2026 at 13:00
arXiv:2603.03485v1 Announce Type: cross Abstract: Recent video diffusion models have achieved impressive capabilities as large-scale generative world models. However, these models often struggle with fine-grained physical consistency, exhibiting physically implausible dynamics over time. In this work, we present \textbf{Phys4D}, a pipeline for learning physics-consistent 4D world representations from video diffusion models. Phys4D adopts \textbf{a three-stage training paradigm} that progressively lifts appearance-driven video diffusion models into physics-consistent 4D world representations. We first bootstrap robust geometry and motion representations through large-scale pseudo-supervised pretraining, establishing a foundation for 4D scene modeling. We then perform physics-grounded supervised fine-tuning using simulation-generated data, enforcing temporally consistent 4D dynamics. Finally, we apply simulation-grounded reinforcement learning to correct residual physical violations that are difficult to capture through explicit supervision. To evaluate fine-grained physical consistency beyond appearance-based metrics, we introduce a set of \textbf{4D world consistency evaluation} that probe geometric coherence, motion stability, and long-horizon physical plausibility. Experimental results demonstrate that Phys4D substantially improves fine-grained spatiotemporal and physical consistency compared to appearance-driven baselines, while maintaining strong generative performance. Our project page is available at https://sensational-brioche-7657e7.netlify.app/

PhyPrompt: RL-based Prompt Refinement for Physically Plausible Text-to-Video Generation

✇cs.AI, q-bio.NC updates on arXiv.org
By: Shang Wu · Chenwei Xu · Zhuofan Xia · Weijian Li · Lie Lu · Pranav Maneriker · Fan Du · Manling Li · Han Liu
5 March 2026 at 13:00
arXiv:2603.03505v1 Announce Type: cross Abstract: State-of-the-art text-to-video (T2V) generators frequently violate physical laws despite high visual quality. We show this stems from insufficient physical constraints in prompts rather than model limitations: manually adding physics details reliably produces physically plausible videos, but requires expertise and does not scale. We present PhyPrompt, a two-stage reinforcement learning framework that automatically refines prompts for physically realistic generation. First, we fine-tune a large language model on a physics-focused Chain-of-Thought dataset to integrate principles like object motion and force interactions while preserving user intent. Second, we apply Group Relative Policy Optimization with a dynamic reward curriculum that initially prioritizes semantic fidelity, then progressively shifts toward physical commonsense. This curriculum achieves synergistic optimization: PhyPrompt-7B reaches 40.8\% joint success on VideoPhy2 (8.6pp gain), improving physical commonsense by 11pp (55.8\% to 66.8\%) while simultaneously increasing semantic adherence by 4.4pp (43.4\% to 47.8\%). Remarkably, our curriculum exceeds single-objective training on both metrics, demonstrating compositional prompt discovery beyond conventional multi-objective trade-offs. PhyPrompt outperforms GPT-4o (+3.8\% joint) and DeepSeek-V3 (+2.2\%, 100$\times$ larger) using only 7B parameters. The approach transfers zero-shot across diverse T2V architectures (Lavie, VideoCrafter2, CogVideoX-5B) with up to 16.8\% improvement, establishing that domain-specialized reinforcement learning with compositional curricula surpasses general-purpose scaling for physics-aware generation.

Pretrained Vision-Language-Action Models are Surprisingly Resistant to Forgetting in Continual Learning

✇cs.AI, q-bio.NC updates on arXiv.org
By: Huihan Liu · Changyeon Kim · Bo Liu · Minghuan Liu · Yuke Zhu
5 March 2026 at 13:00
arXiv:2603.03818v1 Announce Type: cross Abstract: Continual learning is a long-standing challenge in robot policy learning, where a policy must acquire new skills over time without catastrophically forgetting previously learned ones. While prior work has extensively studied continual learning in relatively small behavior cloning (BC) policy models trained from scratch, its behavior in modern large-scale pretrained Vision-Language-Action (VLA) models remains underexplored. In this work, we found that pretrained VLAs are remarkably resistant to forgetting compared with smaller policy models trained from scratch. Simple Experience Replay (ER) works surprisingly well on VLAs, sometimes achieving zero forgetting even with a small replay data size. Our analysis reveals that pretraining plays a critical role in downstream continual learning performance: large pretrained models mitigate forgetting with a small replay buffer size while maintaining strong forward learning capabilities. Furthermore, we found that VLAs can retain relevant knowledge from prior tasks despite performance degradation during learning new tasks. This knowledge retention enables rapid recovery of seemingly forgotten skills through finetuning. Together, these insights imply that large-scale pretraining fundamentally changes the dynamics of continual learning, enabling models to continually acquire new skills over time with simple replay. Code and more information can be found at https://ut-austin-rpl.github.io/continual-vla

CoDAR: Continuous Diffusion Language Models are More Powerful Than You Think

✇cs.AI, q-bio.NC updates on arXiv.org
By: Junzhe Shen · Jieru Zhao · Ziwei He · Zhouhan Lin
4 March 2026 at 13:00
arXiv:2603.02547v1 Announce Type: cross Abstract: We study why continuous diffusion language models (DLMs) have lagged behind discrete diffusion approaches despite their appealing continuous generative dynamics. Under a controlled token--recovery study, we identify token rounding, the final projection from denoised embeddings to tokens, as a primary bottleneck. Building on these insights, we propose CoDAR (Continuous Diffusion with Contextual AutoRegressive Decoder), a two--stage framework that keeps diffusion entirely continuous in an embedding space while learning a strong, context--conditional discretizer: an autoregressive Transformer decoder that cross--attends to the denoised embedding sequence and performs contextualized rounding to tokens. Experiments on LM1B and OpenWebText demonstrate that CoDAR substantially improves generation quality over latent diffusion and becomes competitive with strong discrete DLMs, while exposing a simple decoder--temperature knob to navigate the fluency--diversity trade off.

xLLM Technical Report

✇cs.AI, q-bio.NC updates on arXiv.org
By: Tongxuan Liu · Tao Peng · Peijun Yang · Xiaoyang Zhao · Xiusheng Lu · Weizhe Huang · Zirui Liu · Xiaoyu Chen · Zhiwei Liang · Jun Xiong · Donghe Jin · Minchao Zhang · Jinrong Guo · Yingxu Deng · Xu Zhang · Xianzhe Dong · Siqi Wang · Siyu Wu · Yu Wu · Zihan Tang · Yuting Zeng · Yanshu Wang · Jinguang Liu · Meng Kang · Menxin Li · Yunlong Wang · Yiming Liu · Xiaolong Ma · Yifan Wang · Yichen Zhang · Jinrun Yin · Keyang Zheng · Jiawei Yin · Jun Zhang · Ziyue Wang · Xiaobo Lin · Liangyu Liu · Liwei Lan · Yang Liu · Chunhua Peng · Han Liu · Songcheng Ren · Xuezhu Wang · Yunheng Shen · Yi Wang · Guyue Liu · Yitao Hu · Hui Chen · Tong Yang · Hailong Yang · Jing Li · Guiguang Ding · Ke Zhang
4 March 2026 at 13:00
arXiv:2510.14686v2 Announce Type: replace-cross Abstract: We introduce xLLM, an intelligent and efficient Large Language Model (LLM) inference framework designed for high-performance, large-scale enterprise-grade serving, with deep optimizations for diverse AI accelerators. To address these challenges, xLLM builds a novel decoupled service-engine architecture. At the service layer, xLLM-Service features an intelligent scheduling module that efficiently processes multimodal requests and co-locates online and offline tasks through unified elastic scheduling to maximize cluster utilization. This module also relies on a workload-adaptive dynamic Prefill-Decode (PD) disaggregation policy and a novel Encode-Prefill-Decode (EPD) disaggregation policy designed for multimodal inputs. Furthermore, it incorporates a distributed architecture to provide global KV Cache management and robust fault-tolerant capabilities for high availability. At the engine layer, xLLM-Engine co-optimizes system and algorithm designs to fully saturate computing resources. This is achieved through comprehensive multi-layer execution pipeline optimizations, an adaptive graph mode and an xTensor memory management. xLLM-Engine also further integrates algorithmic enhancements such as optimized speculative decoding and dynamic EPLB, collectively serving to substantially boost throughput and inference efficiency. Extensive evaluations demonstrate that xLLM delivers significantly superior performance and resource efficiency. Under identical TPOT constraints, xLLM achieves throughput up to 1.7x that of MindIE and 2.2x that of vLLM-Ascend with Qwen-series models, while maintaining an average throughput of 1.7x that of MindIE with Deepseek-series models. xLLM framework is publicly available at https://github.com/jd-opensource/xllm and https://github.com/jd-opensource/xllm-service.
❌