arXivDaily arXiv每日学术速递 周一至周五更新

大厂专区

NVIDIA(英伟达)

2026-02-17 至 2026-02-17 共收录 13
2602.15014 2026-02-17 cs.LG cs.CL

Scaling Beyond Masked Diffusion Language Models

超越掩码扩散语言模型的扩展

Subham Sekhar Sahoo, Jean-Marie Lemercier, Zhihan Yang, Justin Deschenaux, Jingyu Liu, John Thickstun, Ante Jukic

机构 * Department of Computer Science, Cornell Tech, NYC, USA(康奈尔科技学院计算机科学系) Department of Computer Science, Cornell University, Ithaca, USA(康奈尔大学计算机科学系) School of Computer and Communication Sciences, EPFL Lausanne, Switzerland(洛桑联邦理工学院计算机与通信科学学院) Department of Computer Science, University of Chicago, Illinois(芝加哥大学计算机科学系) NVIDIA, Santa Clara, USA(英伟达)

AI总结 本文研究了统一状态和插值离散扩散方法的扩展定律,发现掩码扩散在FLOPs效率上可提升12%,并展示了统一状态扩散在GSM8K任务中的优越表现

Comments code: https://github.com/s-sahoo/scaling-dllms

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.14751 2026-02-17 cs.CV

Depth Completion as Parameter-Efficient Test-Time Adaptation

深度补全作为参数高效的测试时适应

Bingxin Ke, Qunjie Zhou, Jiahui Huang, Xuanchi Ren, Tianchang Shen, Konrad Schindler, Laura Leal-Taixé, Shengyu Huang

机构 * NVIDIA ETH Zürich(苏黎世联邦理工学院)

AI总结 CAPA通过参数高效微调方法,在测试时适应预训练的3D基础模型以实现深度补全,提升鲁棒性和多帧一致性。

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.16117 2026-02-17 cs.LG cs.AI cs.CV

DiffusionNFT: Online Diffusion Reinforcement with Forward Process

DiffusionNFT: 在线扩散强化学习与正向过程

Kaiwen Zheng, Huayu Chen, Haotian Ye, Haoxiang Wang, Qinsheng Zhang, Kai Jiang, Hang Su, Stefano Ermon, Jun Zhu, Ming-Yu Liu

机构 * Tsinghua University(清华大学) NVIDIA Stanford University(斯坦福大学)

AI总结 DiffusionNFT通过正向过程优化扩散模型,结合正负生成对比提升强化学习效率,无需CFG且比FlowGRPO更高效。

Comments ICLR 2026 Oral

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.14293 2026-02-17 cs.LG cs.AI

KernelBlaster: Continual Cross-Task CUDA Optimization via Memory-Augmented In-Context Reinforcement Learning

KernelBlaster: 通过记忆增强的上下文强化学习实现持续的跨任务CUDA优化

Kris Shengjun Dong, Sahil Modi, Dima Nikiforov, Sana Damani, Edward Lin, Siva Kumar Sastry Hari, Christos Kozyrakis

机构 * NVIDIA University of California, Berkeley(加州大学伯克利分校)

AI总结 KernelBlaster通过记忆增强的上下文强化学习框架提升CUDA代码优化性能,实现跨多个GPU架构世代的高效优化。

Comments 15 pages, 33 pages with appendix

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.14224 2026-02-17 cs.SD cs.CL cs.MM

The Interspeech 2026 Audio Reasoning Challenge: Evaluating Reasoning Process Quality for Audio Reasoning Models and Agents

Interspeech 2026音频推理挑战:评估音频推理模型和代理的推理过程质量

Ziyang Ma, Ruiyang Xu, Yinghao Ma, Chao-Han Huck Yang, Bohan Li, Jaeyeon Kim, Jin Xu, Jinyu Li, Carlos Busso, Kai Yu, Eng Siong Chng, Xie Chen

机构 * Shanghai Jiao Tong University(上海交通大学) Nanyang Technological University(南洋理工大学) Queen Mary University of London(伦敦大学Queen Mary) NVIDIA(NVIDIA公司) Carnegie Mellon University(卡内基梅隆大学) Qwen Team, Alibaba Group(通义实验室,阿里巴巴集团) Microsoft Corporation(微软公司)

AI总结 Interspeech 2026音频推理挑战通过评估推理过程质量,探讨了音频推理模型和代理在事实性和逻辑性方面的表现及改进方向。

Comments The official website of the Audio Reasoning Challenge: https://audio-reasoning-challenge.github.io

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.13953 2026-02-17 cs.LG

QuRL: Efficient Reinforcement Learning with Quantized Rollout

QuRL: 量化回滚的高效强化学习

Yuhang Li, Reena Elangovan, Xin Dong, Priyadarshini Panda, Brucek Khailany

机构 * NVIDIA Research(NVIDIA研究)

AI总结 QuRL通过量化动作器和自适应截断范围技术,显著提升强化学习回滚效率,实现20%-80%的加速

Comments Accepted to ICLR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.06130 2026-02-17 cs.LG cs.AI cs.CL

Self-Improving World Modelling with Latent Actions

具有潜在动作的自我改进世界建模

Yifu Qiu, Zheng Zhao, Waylon Li, Yftah Ziser, Anna Korhonen, Shay B. Cohen, Edoardo M. Ponti

机构 * University of Edinburgh(爱丁堡大学) Nvidia Research(Nvidia研究) University of Groningen(格罗宁根大学) University of Cambridge(剑桥大学)

AI总结 SWIRL通过将动作视为潜在变量,结合前向世界建模和逆动态建模,实现了对LLM和VLM的自我改进世界建模,在多个基准测试中取得了显著提升。

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.08431 2026-02-17 cs.CV cs.LG

Large Scale Diffusion Distillation via Score-Regularized Continuous-Time Consistency

大规模扩散蒸馏的连续时间一致性 via 分数正则化

Kaiwen Zheng, Yuji Wang, Qianli Ma, Huayu Chen, Jintao Zhang, Yogesh Balaji, Jianfei Chen, Ming-Yu Liu, Jun Zhu, Qinsheng Zhang

机构 * Dept. of Comp. Sci. & Tech., BNRist Center, THU-Bosch ML Center, AI Institute, Tsinghua(清华大学计算机科学与技术系,BNRist中心,THU-Bosch机器学习中心,人工智能研究院) NVIDIA

AI总结 本文提出分数正则化的连续时间一致性模型(rCM),通过引入分数蒸馏作为长跳正则化器,改进了sCM在细节生成和多样性方面的性能,适用于大规模图像和视频扩散任务。

Comments ICLR 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.23519 2026-02-17 cs.CR cs.AI

ReliabilityRAG: Effective and Provably Robust Defense for RAG-based Web-Search

ReliabilityRAG: 有效且可证明的防御方法用于基于检索的Web搜索

Zeyu Shen, Basileal Imana, Tong Wu, Chong Xiang, Prateek Mittal, Aleksandra Korolova

机构 * Department of Computer Science(计算机科学系) Princeton University(普林斯顿大学) Center for Information Technology Policy(信息政策中心) Department of Electrical and Computer Engineering(电气与计算机工程系) NVIDIA(英伟达)

AI总结 ReliabilityRAG通过利用文档可靠性信息,提供更有效且可证明鲁棒的防御方法,以增强基于检索的Web搜索系统对对抗攻击的抵御能力。

Comments Accepted to NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.18053 2026-02-17 cs.RO

V2V-GoT: Vehicle-to-Vehicle Cooperative Autonomous Driving with Multimodal Large Language Models and Graph-of-Thoughts

V2V-GoT: 基于多模态大语言模型和思维图的车对车协同自动驾驶

Hsu-kuang Chiu, Ryo Hachiuma, Chien-Yi Wang, Yu-Chiang Frank Wang, Min-Hung Chen, Stephen F. Smith

机构 * NVIDIA Carnegie Mellon University(卡内基梅隆大学)

AI总结 本文提出V2V-GoT框架,结合多模态大语言模型和图-思维方法,提升车对车协同自动驾驶的感知、预测和规划能力。

Comments Accepted by ICRA 2026 (IEEE International Conference on Robotics and Automation). Project: https://eddyhkchiu.github.io/v2vgot.github.io/ Code: https://github.com/eddyhkchiu/V2V-GoT Dataset: https://huggingface.co/datasets/eddyhkchiu/V2V-GoT-QA

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.09980 2026-02-17 cs.CV cs.RO

V2V-LLM: Vehicle-to-Vehicle Cooperative Autonomous Driving with Multimodal Large Language Models

V2V-LLM:基于多模态大语言模型的车与车协同自动驾驶

Hsu-kuang Chiu, Ryo Hachiuma, Chien-Yi Wang, Stephen F. Smith, Yu-Chiang Frank Wang, Min-Hung Chen

机构 * NVIDIA Carnegie Mellon University(卡内基梅隆大学)

AI总结 本文提出基于多模态大语言模型的V2V-LLM,通过车与车协同感知提升自动驾驶安全性和性能。

Comments Accepted by ICRA 2026 (IEEE International Conference on Robotics and Automation). Project: https://eddyhkchiu.github.io/v2vllm.github.io/ Code: https://github.com/eddyhkchiu/V2V-LLM Dataset: https://huggingface.co/datasets/eddyhkchiu/V2V-GoT-QA

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.10645 2026-02-17 cs.CL

ParoQuant: Pairwise Rotation Quantization for Efficient Reasoning LLM Inference

ParoQuant: 基于成对旋转的量化方法用于高效推理大语言模型推理

Yesheng Liang, Haisheng Chen, Zihan Zhang, Song Han, Zhijian Liu

机构 * NVIDIA MIT(麻省理工学院) UC San Diego(加州大学圣地亚哥分校)

AI总结 ParoQuant通过结合硬件高效旋转与通道缩放,有效解决推理LLMs中的异常值问题,实现更高的准确性和更低的开销。

Comments ICLR 2026 | Project page: https://paroquant.z-lab.ai | GitHub: https://github.com/z-lab/paroquant

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.05376 2026-02-17 cs.LG

LO-BCQ: Block Clustered Quantization for 4-bit (W4A4) LLM Inference

LO-BCQ:用于4位(W4A4)大语言模型推理的块聚类量化

Reena Elangovan, Charbel Sakr, Anand Raghunathan, Brucek Khailany

机构 * NVIDIA Corporation(NVIDIA公司) Department of ECE(电子工程系) Purdue University(普渡大学)

AI总结 LO-BCQ通过块聚类量化方法,在W4A4格式下实现4位大语言模型推理,取得<1%的精度损失,提升推理效率。

Journal ref Transactions on Machine Learning Research, 2025

详情

展开后加载摘要…

URL PDF HTML 收藏