NeurIPS

RotaryQuant: Fitting 120B MoE Models on Consumer Hardware via Fused Compressed-Space Attention

RotaryQuant：通过融合压缩空间注意力在消费级硬件上运行1200亿参数的混合专家模型

Anthony. Lui, Mohamed. Elsaied, N. P. Savani

AI总结 RotaryQuant通过三轴压缩系统及IsoQuant等技术，在消费级硬件上实现1200亿参数MoE模型的高效运行，内存占用低且性能损失极小。

Comments 9 Pages, 9 Tables, Initial submission into NeurIPS 2026

HJ-Gauss: A Monte-Carlo HJ Reachability Scheme

HJ-Gauss: 一种蒙特卡洛HJ可达性方案

Lekan Molu, Venkatraman Renganathan, Namhoon Cho

AI总结 本文提出了一种基于局部PDE线性化的方法，通过冻结系数采样方案解决高维系统中经典网格基HJ求解器内存消耗大的问题，实现了存储和网格无关的算法，适用于高维可达性分析。

Comments NeurIPS 2026 Submission

Fixing It in Post: A Comparative Study of LLM Post-Training Data Quality and Model Performance

事后修复：对LLM事后训练数据质量和模型性能的比较研究

Aladin Djuhera, Swanand Ravindra Kadhe, Syed Zawad, Farhan Ahmed, Heiko Ludwig, Holger Boche

机构 * Technical University Munich（慕尼黑技术大学） ； IBM Research（IBM研究院） 纠错

AI总结 本文通过对比两个开源事后训练数据集，提出了一种系统化编纂方法，生成性能更优的TuluTalk数据集，提升模型表现。

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

Self-Supervised Learning of Graph Representations for Network Intrusion Detection

图表示学习用于网络入侵检测的自监督学习

Lorenzo Guerra, Thomas Chapuis, Guillaume Duc, Pavlo Mozharovskyi, Van-Tam Nguyen

AI总结 GraphIDS通过自监督学习统一图表示学习与异常检测，利用掩码自动编码器和Transformer架构，在网络入侵检测中实现高精度性能。

Comments Accepted at NeurIPS 2025

Journal ref Advances in Neural Information Processing Systems 38 (NeurIPS 2025)

Superposition Yields Robust Neural Scaling

叠加产生稳健的神经扩展

Yizhou Liu, Ziming Liu, Jeff Gore

机构 * Massachusetts Institute of Technology（麻省理工学院） 纠错

AI总结 研究发现表示叠加是神经扩展定律的核心驱动因素，揭示了损失与模型规模之间的反比关系。

Comments Best Paper Runner-up at NeurIPS 2025

Journal ref Advances in Neural Information Processing Systems 38 (2025) 159269--159305

Slow Transition to Low-Dimensional Chaos in Heavy-Tailed Recurrent Neural Networks

Yi Xie, Stefan Mihalas, Łukasz Kuśmierz

Journal ref Advances in Neural Information Processing Systems 38, 136401-136437 (2025)

EEG Foundation Challenge: From Cross-Task to Cross-Subject EEG Decoding

脑电（EEG）基础挑战赛：从跨任务到跨主体的脑电解码

Bruno Aristimunha, Dung Truong, Pierre Guetschel, Seyed Yahya Shirazi, Isabelle Guyon, Alexandre R. Franco, Michael P. Milham, Aviv Dotan, Scott Makeig, Alexandre Gramfort, Jean-Remi King, Marie-Constance Corsi, Pedro A. Valdés-Sosa, Amit Majumdar, Alan Evans, Terrence J Sejnowski, Oren Shriki, Sylvain Chevallier, Arnaud Delorme

AI总结 本文介绍一项包含跨任务跨主体脑电解码、精神病理因素预测两项挑战的大规模脑电竞赛，提供对应基线模型，旨在推动泛化性脑电解码模型及相关临床应用的发展。

Comments Approved at Neurips Competition track. webpage: https://eeg2025.github.io/

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