arXivDaily arXiv每日学术速递 周一至周五更新

期刊&会议

International Conference on Learning Representations · 会议 · Machine Learning

2026-06-17 至 2026-06-17 共收录 8
2606.18153 2026-06-17 cs.CV 新提交

Neural Tree Reconstruction for the Open Forest Observatory

开放森林观测站的神经树重建

Marissa Ramirez de Chanlatte, Arjun Rewari, Trevor Darrell, Derek J. N. Young

机构 * Berkeley AI Research, University of California, Berkeley(加州大学伯克利分校伯克利人工智能研究) Department of Plant Sciences, University of California, Davis(加州大学戴维斯分校植物科学系)

AI总结 针对开放森林观测站中经典运动恢复结构方法重建质量差的问题,提出引入神经辐射场提升3D树重建的细节与鲁棒性,并展望未来工作。

Comments Published as a workshop paper at "Tackling Climate Change with Machine Learning", ICLR 2024

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.17659 2026-06-17 cs.LG 新提交

Physics-Constrained Neural Networks for Improved Short-Term Weather Forecasting: A Case Study over the South Pacific

物理约束神经网络改进短期天气预报:南太平洋案例研究

Egor Bugaev, Fedor Buzaev, Dmitry Efremenko, Denis Derkach, Fedor Ratnikov

机构 * Faculty of Computer Science, Higher School of Economics(高等经济学院计算机科学系)

AI总结 提出三种改进物理约束神经网络(PCNN)的方法,包括升级数值求解器、统一自回归混合块和集成两种神经骨干,在WeatherBench南太平洋子集上相比纯神经网络模型在1-12小时预报中均方根误差降低8-22%,同时保持物理一致性。

Comments Presented at ICLR 2026 Workshop AI and PDE

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.17577 2026-06-17 cs.AI 新提交

Surrogate Assisted Pedestrian Protection Design via a Foundation Model Orchestrated Workflow

基于基础模型编排工作流的代理辅助行人保护设计

Osamu Ito, Akihiko Katagiri, Yoshikazu Nakagawa, Shin Saeki, Jun Shiraishi, Masato Sasaki

机构 * Honda Motor Co., Ltd.(本田汽车有限公司)

AI总结 提出首个基础模型编排的碰撞安全设计工作流,集成代理模型、多目标进化搜索、几何生成器和自然语言接口,将行人保护评估时间从数小时降至秒级。

Journal ref ICLR 2026 Workshop The 2nd Workshop on Foundation Models for Science

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.17389 2026-06-17 cs.CV cs.AI cs.CL cs.LG 新提交

Visuals Lie, Consistency Speaks: Disentangling Spatial Attention from Reliability in Vision-Language Models

视觉会撒谎,一致性说话:在视觉-语言模型中解耦空间注意力与可靠性

Logan Mann, Yi Xia, Ajit Saravanan, Ishan Dave, Saadullah Ismail, Shikhar Shiromani, Emily Huang, Ruizhe Li, Kevin Zhu

机构 * University of California, Santa Barbara(加州大学圣塔芭芭拉分校) Algoverse AI Research(Algoverse AI研究) University of California, Berkeley(加州大学伯克利分校)

AI总结 本文提出VLM可靠性探针(VRP),通过结构注意力指标和生成动态分析,发现空间注意力与准确性几乎无关(R≈0.001),而自一致性是可靠性的主要预测因子(R=0.429),揭示了视觉特征与最终生成之间的符号脱离现象。

Comments 16 pages. Accepted to the ICLR 2026 Workshop on Multimodal Intelligence. Code: https://github.com/itsloganmann/VLM-Reliability-Probe

详情

展开后加载摘要…

URL PDF HTML 收藏
2606.17312 2026-06-17 cs.AI 新提交

Quantifying Consistency in LLM Logical Reasoning via Structural Uncertainty

通过结构不确定性量化LLM逻辑推理中的一致性

Baishali Chaudhury, Mengdie Flora Wang, Hyunji Hayley Park, Rahul Ghosh, Sungmin Hong, Jae Oh Woo

机构 * AWS Generative AI Innovation Center(AWS生成式AI创新中心)

AI总结 提出结构不确定性框架,通过自偏好排序的稳定性评估LLM推理一致性,在逻辑和数学任务中与答案分散度互补,提升不可靠实例识别。

Comments Published at ICLR 2026 Workshop on Logical Reasoning of Large Language Models. Accepted as best paper

详情

展开后加载摘要…

URL PDF HTML 收藏
2604.10827 2026-06-17 cs.AI 版本更新

Know Thy Reasoner: Not All Language Models Explore Alike

你的模型多样性,而非方法,决定推理策略

Moulik Choraria, Argyrios Gerogiannis, Anirban Das, Supriyo Chakraborty, Sourya Basu, Sambit Sahu, Lav R. Varshney

机构 * UIUC(伊利诺伊大学香槟分校) Capital One

AI总结 本文提出模型多样性影响推理策略,通过理论框架分析推理不确定性,验证了不同模型在深度精炼和并行采样中的表现差异。

Comments This is a full-length extension of the workshop paper that appeared in the ICLR 2026 Workshop on LLM Reasoning

详情

展开后加载摘要…

URL PDF HTML 收藏
2508.04492 2026-06-17 cs.CV cs.AI

Learning Robust Intervention Representations with Delta Embeddings

通过delta嵌入学习鲁棒的干预表示

Panagiotis Alimisis, Christos Diou

机构 * Department of Informatics and Telematics(信息与电信学系)

AI总结 本文提出通过潜在空间中的可操作反事实表示提升模型鲁棒性,提出因果delta嵌入方法,在无需额外监督的情况下学习因果表示,实验显示其在合成和现实基准中表现优异。

Comments ICLR 2026, Poster

Journal ref International Conference on Learning Representations (ICLR), 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.21315 2026-06-17 cs.LG cs.CV stat.ML 版本更新

Does the Data Processing Inequality Reflect Practice? On the Utility of Low-Level Tasks

数据处理不等式是否反映实践?论低级任务的有用性

Roy Turgeman, Tom Tirer

机构 * Faculty of Engineering Bar-Ilan University(巴伊兰大学工程学院)

AI总结 本文研究低级处理(如去噪、编码)如何提升分类性能,证明在有限样本下存在预处理可提高准确率,并通过实验验证理论趋势。

Comments ICLR 2026 (camera-ready). Code is available at: https://github.com/serveroy/process-before-you-classify

Journal ref The Fourteenth International Conference on Learning Representations (ICLR 2026)

详情

展开后加载摘要…

URL PDF HTML 收藏