arXivDaily arXiv每日学术速递 周一至周五更新

期刊&会议

NeurIPS

Conference on Neural Information Processing Systems · 会议 · Machine Learning

共收录 17318
2409.08439 2026-03-10 cs.RO cs.AI cs.LG cs.SY eess.SY

Input-to-State Stable Coupled Oscillator Networks for Closed-form Model-based Control in Latent Space

用于潜在空间模型基于控制的输入到状态稳定的耦合振荡器网络

Maximilian Stölzle, Cosimo Della Santina

AI总结 本文提出了一种耦合振荡器网络模型,通过引入拉格朗日结构和可逆映射,实现了潜在空间中高效的模型基于控制。

Comments 38th Conference on Neural Information Processing Systems (NeurIPS 2024) spotlight, 50 pages

Journal ref Stölzle, Maximilian, and Cosimo Della Santina. "Input-to-state stable coupled oscillator networks for closed-form model-based control in latent space." Advances in Neural Information Processing Systems 37 (2024): 82010-82059

详情

展开后加载摘要…

URL PDF HTML 收藏
2408.15205 2026-03-10 cs.CV

Leveraging Hallucinations to Reduce Manual Prompt Dependency in Promptable Segmentation

利用幻觉减少可提示分割中的手动提示依赖

Jian Hu, Jiayi Lin, Junchi Yan, Shaogang Gong

机构 * Queen Mary University of London(伦敦大学玛丽女王学院) Shanghai Jiao Tong University(上海交通大学)

AI总结 本文提出ProMaC框架,通过利用MLLM幻觉挖掘任务相关信息,提升分割精度与遮罩生成效果。

Comments NeurIPS 2024

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.07614 2026-03-10 cs.CV

Looking Into the Water by Unsupervised Learning of the Surface Shape

通过无监督学习探究水面形状

Ori Lifschitz, Tali Treibitz, Dan Rosenbaum

机构 * Hatter Department of Marine Technologies(海洋技术系) Charney School of Marine Sciences, University of Haifa(海洋科学学院,海法大学)

AI总结 本文提出一种基于神经场网络的无监督学习方法,用于从空中去除水面折射引起的图像失真,并有效估计水面形状。

Journal ref Published The Thirty-ninth Annual Conference on Neural Information Processing Systems (NeurIPS) 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.07368 2026-03-10 cs.CL cs.AI

Position: LLMs Must Use Functor-Based and RAG-Driven Bias Mitigation for Fairness

位置:LLMs必须使用基于函子和RAG驱动的偏见缓解以实现公平性

Ravi Ranjan, Utkarsh Grover, Agorista Polyzou

机构 * Knight Foundation School of Computing and Information Sciences(骑士基金会计算与信息科学学院) Florida International University(佛罗里达国际大学) College of Engineering(工程学院) University of South Florida(佛罗里达州立大学)

AI总结 本文提出通过范畴论转换和RAG结合的方法,解决LLMs中的偏见问题,以实现公平性。

Comments 24 pages, 3 figures

Journal ref Review available from NeurIPS 2025 reviwers

详情

展开后加载摘要…

URL PDF HTML 收藏
2506.05587 2026-03-10 cs.AI cs.CL cs.DB cs.LG

MMTU: A Massive Multi-Task Table Understanding and Reasoning Benchmark

MMTU: 一个大规模多任务表格理解和推理基准

Junjie Xing, Yeye He, Mengyu Zhou, Haoyu Dong, Shi Han, Lingjiao Chen, Dongmei Zhang, Surajit Chaudhuri, H. V. Jagadish

机构 * University of Michigan(密歇根大学) Microsoft Corporation(微软公司)

AI总结 MMTU是一个大规模多任务表格理解和推理基准,旨在评估模型在专家级别处理真实表格的能力,揭示了当前模型在表格理解、推理和编码方面的挑战。

Comments Full version of a paper accepted at NeurIPS 2025; Code and data available at https://github.com/MMTU-Benchmark/MMTU and https://huggingface.co/datasets/MMTU-benchmark/MMTU

详情

展开后加载摘要…

URL PDF HTML 收藏
2505.14996 2026-03-10 cs.CL cs.AI cs.LG

MAS-ZERO: Designing Multi-Agent Systems with Zero Supervision

MAS-ZERO:无需监督设计多智能体系统

Zixuan Ke, Austin Xu, Yifei Ming, Xuan-Phi Nguyen, Ryan Chin, Caiming Xiong, Shafiq Joty

机构 * Salesforce AI Research(Salesforce AI研究院) Massachusetts Institute of Technology(麻省理工学院)

AI总结 MAS-ZERO通过元层面设计实现无需监督的多智能体系统自动设计,提升推理、编程和代理任务的性能。

Comments SEA@NeurIPS (Oral) 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.07162 2026-03-10 cs.LG

Spectral Conditioning of Attention Improves Transformer Performance

注意力的谱条件化提升变换器性能

Hemanth Saratchandran, Simon Lucey

机构 * Australian Institute for Machine Learning(澳大利亚机器学习研究所)

AI总结 通过优化注意力层的谱特性以降低雅可比矩阵的条件数,提升变换器模型的性能。

Comments NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.07006 2026-03-10 cs.AR

Mozart: Modularized and Efficient MoE Training on 3.5D Wafer-Scale Chiplet Architectures

Mozart:模块化和高效的3.5D晶圆级芯片组架构上的MoE训练

Shuqing Luo, Ye Han, Pingzhi Li, Jiayin Qin, Jie Peng, Yang, Zhao, Yu, Cao, Tianlong Chen

AI总结 Mozart通过模块化和高效的算法-硬件协同设计,提升3.5D晶圆级芯片组架构上MoE模型的训练效率与资源利用率。

Comments NeurIPS 2025 Spotlight

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.06894 2026-03-10 cs.LG cs.CV

Learning From Design Procedure To Generate CAD Programs for Data Augmentation

从设计过程学习生成用于数据增强的CAD程序

Yan-Ying Chen, Dule Shu, Matthew Hong, Andrew Taber, Jonathan Li, Matthew Klenk

机构 * Toyota Research Institute(丰田研究院)

AI总结 本文提出了一种基于设计过程的CAD程序生成方法,通过引入有机形状和样条基 curvature 提高几何多样性,以增强数据增强效果。

Comments Accepted by NeurIPS 2025 Workshop: Deep Learning for Code in the Agentic Era

详情

展开后加载摘要…

URL PDF HTML 收藏
2508.04016 2026-03-10 cs.CV

S$^2$Q-VDiT: Accurate Quantized Video Diffusion Transformer with Salient Data and Sparse Token Distillation

S$^2$Q-VDiT: 精确量化视频扩散变换器与显著数据和稀疏令牌蒸馏

Weilun Feng, Haotong Qin, Chuanguang Yang, Xiangqi Li, Han Yang, Yuqi Li, Zhulin An, Libo Huang, Michele Magno, Yongjun Xu

机构 * State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences(人工智能安全国家重点实验室,计算技术研究所,中国科学院) University of Chinese Academy of Sciences(中国科学院大学) ETH Zürich(苏黎世联邦理工学院)

AI总结 S$^2$Q-VDiT通过显著数据选择和稀疏令牌蒸馏提升视频扩散变换器的量化性能,实现无损压缩和加速推理。

Comments Accepted by NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.02197 2026-03-10 cs.LG cs.AI cs.SI

An Efficient Local Search Approach for Polarized Community Discovery in Signed Networks

在带符号网络中发现极化社区的高效局部搜索方法

Linus Aronsson, Morteza Haghir Chehreghani

机构 * Chalmers University of Technology & University of Gothenburg(楚德斯技术大学及哥德堡大学)

AI总结 本文提出了一种在带符号网络中发现极化社区的高效局部搜索方法,解决了以往方法中解决方案大小不平衡的问题,并在大规模网络上实现了线性收敛率。

Journal ref The Thirty-Ninth Annual Conference on Neural Information Processing Systems (NeurIPS 2025)

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.06142 2026-03-09 cs.LG cond-mat.dis-nn cs.AI cs.NE stat.ML

Predictive Coding Graphs are a Superset of Feedforward Neural Networks

预测编码图是前馈神经网络的超集

Björn van Zwol

机构 * Department of Information & Computing Sciences, Utrecht University(信息与计算科学系,乌特勒支大学)

AI总结 预测编码图(PCGs)作为前馈神经网络的超集,为机器学习提供了新的视角和方法。

Comments 11 pages, 3 figures. Accepted at the NeuroAI Workshop @ NeurIPS 2024. OpenReview: https://openreview.net/forum?id=J36z3R0sNq

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.05664 2026-03-09 cs.LG

KLASS: KL-Guided Fast Inference in Masked Diffusion Models

KLASS: 在掩码扩散模型中基于KL的快速推理

Seo Hyun Kim, Sunwoo Hong, Hojung Jung, Youngrok Park, Se-Young Yun

机构 * KAIST AI(韩国科学技术院人工智能学院)

AI总结 KLASS通过利用令牌级KL散度实现快速稳定生成,显著提升扩散模型推理速度并保持高质量输出。

Comments NeurIPS 2025 Spotlight. Code: https://github.com/shkim0116/KLASS

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.17859 2026-03-09 eess.SY cs.LG cs.SY

Mixed Monotonicity Reachability Analysis of Neural ODE: A Trade-Off Between Tightness and Efficiency

神经ODE的混合单调性可达性分析:紧致性与效率之间的权衡

Abdelrahman Sayed Sayed, Pierre-Jean Meyer, Mohamed Ghazel

AI总结 本文提出了一种基于区间的方法,利用混合单调性技术对神经ODE进行可达性分析,在紧致性和效率之间取得平衡,适用于高维实时安全应用。

Comments 27 pages, 11 figures, Accepted for publication in PMLR proceedings of NeurReps 2025 co-located with NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.19674 2026-03-09 cs.LG cs.CV

C^2Prompt: Class-aware Client Knowledge Interaction for Federated Continual Learning

C²Prompt:面向联邦持续学习的类感知客户端知识交互

Kunlun Xu, Yibo Feng, Jiangmeng Li, Yongsheng Qi, Jiahuan Zhou

机构 * Wangxuan Institute of Computer Technology Peking University(北京大学王轩计算机技术研究所) University of Chinese Academy of Sciences(中国科学院大学) Inner Mongolia University of Technology(内蒙古科技大学)

AI总结 C²Prompt通过增强类层面知识一致性,解决联邦持续学习中的时间与空间遗忘问题,实现最先进的性能。

Comments Accepted by NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2507.06543 2026-03-09 cs.CV

Token Bottleneck: One Token to Remember Dynamics

令牌瓶颈:一个令牌来记住动态

Taekyung Kim, Dongyoon Han, Byeongho Heo, Jeongeun Park, Sangdoo Yun

机构 * NAVER AI Lab(NAVER AI实验室) Korea University(韩国大学)

AI总结 Token Bottleneck通过压缩场景为一个令牌并利用补丁提示来学习动态场景的序列表示,从而在视觉跟踪和机器人操作等任务中实现优越性能。

Comments NeurIPS 2025, 18 pages, 9 figures, 10 tables, project page: https://token-bottleneck.github.io, code: https://github.com/naver-ai/tobo

详情

展开后加载摘要…

URL PDF HTML 收藏
2505.19297 2026-03-09 cs.CV

Alchemist: Turning Public Text-to-Image Data into Generative Gold

炼金术:将公共文本到图像数据转化为生成黄金

Valerii Startsev, Alexander Ustyuzhanin, Alexey Kirillov, Dmitry Baranchuk, Sergey Kastryulin

机构 * Yandex Research(Yandex研究院) HSE(莫斯科大学) Yandex MSU(莫斯科大学)

AI总结 本文提出了一种利用预训练模型生成高质量通用SFT数据集的方法,通过Alchemist数据集提升了文本到图像模型的生成质量并公开了微调权重。

Comments Accepted to the Datasets and Benchmarks Track of the 39th Conference on Neural Information Processing Systems

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.17721 2026-03-09 cs.LG cs.AI cs.MA

Aligning Compound AI Systems via System-level DPO

通过系统级DPO对复合AI系统进行对齐

Xiangwen Wang, Yibo Jacky Zhang, Zhoujie Ding, Katherine Tsai, Haolun Wu, Sanmi Koyejo

机构 * Stanford University(斯坦福大学) University of Illinois Urbana Champaign(伊利诺伊大学厄巴纳-香槟分校) Mila Quebec AI Institute(魁北克AI研究院)

AI总结 本文提出SysDPO框架,通过系统级DPO实现复合AI系统的联合对齐,解决了组件间非可微分交互和系统偏好转换的问题。

Comments NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2412.01711 2026-03-09 cs.CL

Towards Resource Efficient and Interpretable Bias Mitigation in Large Language Models

迈向大型语言模型中资源高效且可解释的偏见缓解

Schrasing Tong, Eliott Zemour, Jessica Lu, Rawisara Lohanimit, Lalana Kagal

机构 * Department of Electrical Engineering and Computer Science(电气工程与计算机科学系) Massachusetts Institute of Technology(麻省理工学院)

AI总结 本文提出了一种通过小型专家模型高效缓解大型语言模型偏见的方法,通过解码时添加去偏信号,减少性别、种族和宗教偏见,同时保持语言模型性能。

Comments 38th Conference on Neural Information Processing Systems (NeurIPS 2024) Safe Generative AI Workshop. Updated results in V2

详情

展开后加载摘要…

URL PDF HTML 收藏
2409.17137 2026-03-09 cs.LG cs.CV

PACE: Marrying generalization in PArameter-efficient fine-tuning with Consistency rEgularization

PACE:将参数高效微调的泛化能力与一致性正则化相结合

Yao Ni, Shan Zhang, Piotr Koniusz

机构 * The Australian National University(澳大利亚国立大学) Data61 CSIRO Australian Institute for Machine Learning, The University of Adelaide(澳大利亚机器学习研究所,阿德莱德大学)

AI总结 PACE通过结合参数高效微调与一致性正则化,提升模型泛化能力与知识保留,优于现有PEFT方法。

Comments Accepted by NeurIPS 2024 as a spotlight

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.24544 2026-03-06 stat.ML cs.LG math.PR

Quantitative convergence of trained single layer neural networks to Gaussian processes

训练单层神经网络向高斯过程定量收敛性

Eloy Mosig, Andrea Agazzi, Dario Trevisan

机构 * Department of Mathematics University of Pisa(数学系,比萨大学) Department of Mathematics and Statistics University of Bern(数学与统计系,伯恩大学)

AI总结 本文研究了训练单层神经网络在无限宽度极限下向高斯过程的定量收敛性,提供了网络输出与高斯近似之间距离的显式上界,并分析了架构参数和训练动态对收敛性的影响。

Comments Submitted and accepted at NeurIPS 2025, main body of 10 pages, 3 figures, 28 pages of supplementary material. Corrected an issue in the proof of Proposition 3.7

详情

展开后加载摘要…

URL PDF HTML 收藏
2507.01785 2026-03-06 cs.CL cs.AI cs.LG

MuRating: A High Quality Data Selecting Approach to Multilingual Large Language Model Pretraining

MuRating: 一种用于多语言大语言模型预训练的高质量数据选择方法

Zhixun Chen, Ping Guo, Wenhan Han, Yifan Zhang, Binbin Liu, Haobin Lin, Fengze Liu, Yan Zhao, Bingni Zhang, Taifeng Wang, Yin Zheng, Trevor Cohn, Meng Fang

机构 * Hong Kong University of Science and Technology (Guangzhou)(香港理工大学(广州)) ByteDance(字节跳动) Eindhoven University of Technology(埃因霍温理工大学) University of Melbourne(墨尔本大学) University of Liverpool(利物浦大学)

AI总结 MuRating通过多语言评估器选择高质量数据,提升多语言大语言模型的预训练效果。

Comments NeurIPS 2025 poster

详情

展开后加载摘要…

URL PDF HTML 收藏
2301.07695 2026-03-06 cs.CL cs.AI

EHRSQL: A Practical Text-to-SQL Benchmark for Electronic Health Records

EHRSQL:一个用于电子健康记录的实用文本到SQL基准数据集

Gyubok Lee, Hyeonji Hwang, Seongsu Bae, Yeonsu Kwon, Woncheol Shin, Seongjun Yang, Minjoon Seo, Jong-Yeup Kim, Edward Choi

AI总结 EHRSQL是一个用于电子健康记录的文本到SQL基准数据集,旨在解决医疗场景中的复杂查询和时间表达问题。

Comments Published as a conference paper at NeurIPS 2022 (Track on Datasets and Benchmarks)

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.03682 2026-03-05 eess.IV cs.CV

Polyp Segmentation Using Wavelet-Based Cross-Band Integration for Enhanced Boundary Representation

基于小波域交叉频带整合的多发性息肉分割

Haesung Oh, Jaesung Lee

机构 * Department of Artificial Intelligence, Chung-Ang University(人工智能系, Chung-Ang大学)

AI总结 本文提出基于小波域交叉频带整合的息肉分割方法,通过整合灰度和RGB信息提升边界表示精度,实验表明在多个数据集上效果优于传统方法。

Comments 39th Annual Conference on Neural Information Processing Systems in Europe (EurIPS 2025) Workshop, Copenhagen, Denmark, 2-7 December 2025 MedEurIPS:Medical Imagine Meets EurIPS

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.08269 2026-03-05 cs.CV

Re-coding for Uncertainties: Edge-awareness Semantic Concordance for Resilient Event-RGB Segmentation

不确定性重编码:面向鲁棒事件-RGB分割的边缘感知语义一致性

Nan Bao, Yifan Zhao, Lin Zhu, Jia Li

机构 * State Key Laboratory of Virtual Reality Technology and Systems, SCSE & QRI, Beihang University(虚拟现实技术与系统国家重点实验室,SCSE与QRI,北京航空航天大学) School of Computer Science and Technology, Beijing Institute of Technology(计算机科学与技术学院,北京理工大学)

AI总结 本文提出边缘感知语义一致性框架,通过重编码和不确定性优化实现鲁棒的事件-RGB分割,提升极端条件下的分割性能。

Comments Accepted to NeurIPS 2025; code and datasets available at https://github.com/iCVTEAM/ESC

详情

展开后加载摘要…

URL PDF HTML 收藏
2505.16985 2026-03-05 cs.CV cs.AI cs.LG cs.RO

Extremely Simple Multimodal Outlier Synthesis for Out-of-Distribution Detection and Segmentation

极简多模态异常合成用于分布外检测与分割

Moru Liu, Hao Dong, Jessica Kelly, Olga Fink, Mario Trapp

机构 * Technical University of Munich(慕尼黑技术大学) ETH Zürich(苏黎世联邦理工学院) Fraunhofer IKS(弗劳恩霍夫研究所) EPFL(苏黎世联邦理工学院)

AI总结 本文提出极简多模态异常合成方法Feature Mixing,通过理论支持提升OOD检测性能,实验表明其在多个数据集上达到最优效果,速度提升显著。

Comments NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2505.10118 2026-03-05 cs.CV cs.CL

Why 1 + 1 < 1 in Visual Token Pruning: Beyond Naive Integration via Multi-Objective Balanced Covering

为何1+1<1在视觉令牌修剪中:通过多目标平衡覆盖超越简单整合

Yangfu Li, Hongjian Zhan, Tianyi Chen, Qi Liu, Yue Lu

机构 * East China Normal University(华东师范大学) Shanghai Jiao Tong University(上海交通大学) Chongqing Institute of East China Normal University(华东师范大学重庆研究院) Shanghai University of Engineering Science(上海工程技术大学)

AI总结 MoB通过多目标平衡覆盖方法,解决视觉令牌修剪中提示对齐与视觉保留的权衡问题,实现高效且高性能的修剪效果。

Comments 31 pages,9 figures,conference

Journal ref Advances in Neural Information Processing Systems 39 (NeurIPS 2025)

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.02846 2026-03-04 cs.LG cs.AI

Learning Memory-Enhanced Improvement Heuristics for Flexible Job Shop Scheduling

学习记忆增强的改进启发式方法用于灵活作业车间调度

Jiaqi Wang, Zhiguang Cao, Peng Zhao, Rui Cao, Yubin Xiao, Yuan Jiang, You Zhou

机构 * Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education(教育部符号计算与知识工程重点实验室) College of Software(软件学院) College of Computer Science and Technology(计算机科学与技术学院) Singapore Management University(新加坡管理大学) Nanyang Technological University(南洋理工大学)

AI总结 本文提出MIStar框架,通过记忆增强的异构图神经网络和并行贪婪搜索策略,提升灵活作业车间调度的调度效果。

Comments 39th Conference on Neural Information Processing Systems (NeurIPS 2025)

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.05116 2026-03-04 cs.LG cs.CV

Value Gradient Guidance for Flow Matching Alignment

价值梯度指导下的流匹配对齐

Zhen Liu, Tim Z. Xiao, Carles Domingo-Enrich, Weiyang Liu, Dinghuai Zhang

机构 * The Chinese University of Hong Kong (Shenzhen)(香港中文大学(深圳)) University of Tübingen(图宾根大学) Microsoft Research(微软研究院) The Chinese University of Hong Kong(香港中文大学) Mila – Quebec AI Institute(魁北克AI研究所)

AI总结 VGG-Flow通过价值梯度指导实现流匹配模型的高效微调与先验保留对齐。

Comments Accepted at NeurIPS 2025; 26 pages, 20 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2407.16893 2026-03-04 cs.CY cs.AI cs.CL

The Price of Prompting: Profiling Energy Use in Large Language Models Inference

提示的成本:大型语言模型推理中能源使用的 profiling

Erik Johannes Husom, Arda Goknil, Lwin Khin Shar, Sagar Sen

机构 * SINTEF Singapore Management University(新加坡管理大学)

AI总结 本文提出MELODI框架,用于分析大型语言模型推理过程中的能源消耗,通过构建数据集研究提示属性与能源支出的关系,为可持续LLM部署提供基础工具。

Comments 11 pages, 5 figures. Submitted to NeurIPS 2024. The released code and dataset are available at https://github.com/ejhusom/MELODI

详情

展开后加载摘要…

URL PDF HTML 收藏