arXivDaily arXiv每日学术速递 周一至周五更新

大厂专区

Tencent(腾讯)

2026-08-26 至 2026-08-26 共收录 5
2608.23566 2026-08-26 cs.LG cs.AI cs.CL 版本更新

Best Practice Critic Optimization

如何稳定且高效地训练评价模型

Penghui Qi, Xiangxin Zhou, Wee Sun Lee

机构 * National University of Singapore(新加坡国立大学) Tencent Hunyuan(腾讯混元)

AI总结 该研究提出BPCO流程稳定高效训练评价模型,在数学推理任务上,其优于基线,且采样1个响应时可匹配或超过基于组的GRPO基线。

详情

展开后加载摘要…

URL PDF HTML 收藏
2608.20804 2026-08-26 cs.CL cs.AI 版本更新

Denoising the Future: Context-Aware Spectral Diffusion for Temporal Knowledge Graph Extrapolation

去噪未来:用于时序知识图谱外推的上下文感知谱扩散

Yanglei Gan, Peng He, Run Lin, Peiyuan Jiang, Yifan Wang, Qiao Liu

机构 * Southwest Minzu University(西南民族大学) University of Electronic Science and Technology of China(电子科技大学) Zhejiang University(浙江大学) Tencent(腾讯)

AI总结 针对现有扩散式时序知识图谱外推方法的目标判别信号削弱问题,提出FreqDiff频率感知扩散框架,通过双流去噪器与频域正则化项提升性能,在四个公开基准上达最优表现。

Comments EMNLP 2026 Main

详情

展开后加载摘要…

URL PDF HTML 收藏
2608.17566 2026-08-26 cs.CV 版本更新

CoinVE-200K: A Large-Scale High-Quality Dataset for Compositional Instruction-Guided Video Editing

CoinVE-200K:用于组合式指令引导视频编辑的大规模高质量数据集

Fuchen Long, Cong Wang, Zitao Gao, Wenhao Zhong, Yu Cheng, Xiaolu Hou, Yan Li, Xiao Cao, Xinlong Sun, Xi Chen, Yu Liu

机构 * Tencent(腾讯)

AI总结 本文提出用于组合式指令引导视频编辑的大规模高质量数据集CoinVE-200K,构建基准CoinVE-Bench及22B参数模型CoinVE-Edit,在基准测试中表现优异。

Comments Project page: this https URL (https://coinve200k.github.io) and Dataset is available at this https URL (https://huggingface.co/datasets/FireCRT/CoinVE-200K) and see source codes at this https URL (https://github.com/coinve200k/CoinVE-200K)

详情

展开后加载摘要…

URL PDF HTML 收藏
2601.16520 2026-08-26 cs.CV cs.AI cs.CL 版本更新

TangramPuzzle: Evaluating Multimodal Large Language Models with Compositional Spatial Reasoning

TangramPuzzle: 通过组合空间推理评估多模态大语言模型

Daixian Liu, Jiayi Kuang, Yinghui Li, Yangning Li, Di Yin, Haoyu Cao, Xing Sun, Ying Shen, Hai-Tao Zheng, Liang Lin, Philip S. Yu

机构 * Tsinghua University(清华大学) Sun-Yat Sen University(孙逸人大学) Tencent Youtu Lab(腾讯优图实验室) University of Illinois Chicago(伊利诺伊大学香槟分校)

AI总结 TangramPuzzle通过几何基准测试评估多模态大语言模型的组合空间推理能力,发现模型在匹配轮廓时忽视几何约束,导致碎片变形。

Comments EMNLP 2026 Findings

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.06848 2026-08-26 quant-ph cs.CC cs.DS cs.LG 版本更新

Reconquering Bell sampling on qudits: stabilizer learning and testing, quantum pseudorandomness bounds, and more

Jonathan Allcock, Joao F. Doriguello, Gábor Ivanyos, Miklos Santha

机构 * Tencent Quantum Laboratory(腾讯量子实验室) HUN-REN Alfréd Rényi Institute of Mathematics(匈牙利冯·诺依曼数学研究所) HUN-REN Institute for Computer Science and Control(匈牙利计算机科学与控制研究所) Centre for Quantum Technologies(量子技术中心) National University of Singapore(新加坡国立大学) CNRS, IRIF, Université Paris Cité(法国国家科学研究中心、IRIF、巴黎Cité大学)

Comments 48 pages, 1 figure. v2: slightly improved results, some proofs and sections rewritten, references added and fixed; v3: a few sections completely rewritten, better stabiliser testing bounds, several typos and inconsistencies fixed

详情

展开后加载摘要…

URL PDF HTML 收藏