arXivDaily arXiv每日学术速递 周一至周五更新

大厂专区

NVIDIA(英伟达)

2026-06-19 至 2026-06-19 共收录 4
2605.31158 2026-06-19 cs.CV cs.LG 版本更新

Light Interaction: Training-Free Inference Acceleration for Interactive Video World Models

光交互:交互式视频世界模型的免训练推理加速

Jiacheng Lu, Haoyi Zhu, Sipei Yi, Enze Xie, Yu Li, Cheng Zhuo

机构 * Zhejiang University(浙江大学) NVIDIA

AI总结 针对交互式视频世界模型推理成本高的问题,提出免训练加速框架Light Interaction,通过自适应上下文管理、去噪缓存加速和3D块稀疏注意力实现最高2.59倍加速。

Comments 13 pages, 6 figures, 3 tables. Project page: https://2843721358l-del.github.io/Light-Interaction-Project/

详情

展开后加载摘要…

URL PDF HTML 收藏
2605.15231 2026-06-19 cs.LG cs.CV 版本更新

Mask-Morph Graph U-Net: A Generalisable Mesh-Based Surrogate for Crashworthiness Field Prediction under Large Geometric Variation

Mask-Morph Graph U-Net:一种通用的基于网格的替代模型,用于在大几何变化下预测碰撞worthiness领域

Haoran Li, Tobias Lehrer, Yingxue Zhao, Haosu Zhou, Philipp Stocker, Tobias Pfaff, Marcus Wagner, Nan Li

机构 * Dyson School of Design Engineering, Imperial College London(帝国理工学院伦敦设计工程学院) TUM School of Engineering and Design, Technical University of Munich(慕尼黑技术大学工程与设计学院) Faculty of Mechanical Engineering, OTH Regensburg(雷根斯堡机械工程学院) NVIDIA(NVIDIA公司)

AI总结 本文提出Mask-Morph Graph U-Net,通过特征对齐的重心参数化和节点掩码预训练,提升网格模拟的通用性和数据效率,适用于碰撞worthiness设计探索。

Comments 48 pages, 15 figures, jounral paper under review

详情

展开后加载摘要…

URL PDF HTML 收藏
2506.06952 2026-06-19 cs.CV 版本更新

LaTtE-Flow: Layerwise Timestep-Expert Flow-based Transformer

LaTtE-Flow: 基于层间时间步专家流的Transformer

Ying Shen, Zhiyang Xu, Jiuhai Chen, Shizhe Diao, Jiaxin Zhang, Yuguang Yao, Joy Rimchala, Ismini Lourentzou, Lifu Huang

机构 * University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校) University of Maryland(马里兰大学) Nvidia(英伟达) Salesforce AI Research(Salesforce AI研究) Intuit AI Research(Intuit AI研究)

AI总结 提出LaTtE-Flow,一种基于预训练视觉语言模型的高效统一架构,通过层间时间步专家流和条件残差注意力机制,实现图像理解与生成,生成速度提升约6倍。

Comments Unified multimodal model, Flow-matching

详情

展开后加载摘要…

URL PDF HTML 收藏
2503.17386 2026-06-19 eess.SY cs.LG cs.SY 版本更新

A graph neural network surrogate model for mesh-based crashworthiness prediction of vehicle panel components

基于图神经网络的网格级车辆面板部件耐撞性预测代理模型

Haoran Li, Yingxue Zhao, Haosu Zhou, Tobias Pfaff, Nan Li

机构 * Dyson School of Design Engineering, Imperial College London(迪森设计工程学院,帝国理工学院伦敦分校) NVIDIA

AI总结 提出递归图U-Net (ReGUNet) 代理模型,通过图表示有限元网格,结合层次架构和递归机制,高效准确预测车辆B柱等面板部件的动态变形和耐撞性指标。

Comments Accepted manuscript version. Final published version available in Results in Engineering via DOI: 10.1016/j.rineng.2026.110925

Journal ref Results in Engineering 30 (2026) 110925

详情

展开后加载摘要…

URL PDF HTML 收藏