NeurIPS

GaLoRA: Parameter-Efficient Graph-Aware LLMs for Node Classification

GaLoRA：参数高效图感知大语言模型用于节点分类

Mayur Choudhary, Saptarshi Sengupta, Katerina Potika

AI总结 GaLoRA通过整合结构信息提升LLMs在文本属性图节点分类中的性能，仅用0.24%的参数量即可达到与最新模型相当的效果。

Comments 10 pages, 2 figures, 11 tables, 39th Conference on Neural Information Processing Systems (NeurIPS 2025) Workshop

MITRA: An AI Assistant for Knowledge Retrieval in Physics Collaborations

MITRA：一种用于物理学协作中知识检索的AI助手

Abhishikth Mallampalli, Sridhara Dasu

AI总结 MITRA是一种基于RAG的AI助手，通过本地化框架实现物理学协作中的高效知识检索与生成。

Comments Accepted at NeurIPS 2025 Machine Learning for the Physical Sciences workshop and Lepton Photon conference 2025 (Computing AI/ML track)

What We Don't C: Manifold Disentanglement for Structured Discovery

我们未捕捉到的：结构发现的流解耦

Brian Rogers, Micah Bowles, Chris J. Lintott, Steve Croft, Oliver N. F. King, James Kostas Ray

AI总结 What We Don't C通过移除条件指导中的信息来解耦潜在子空间，从而提升对潜在表示的分析与再利用能力。

Comments v2: Preprint of extended version. 21 pages. v1: Short version accepted to the Machine Learning and the Physical Sciences workshop at NeurIPS 2025 (Number 315: https://ml4physicalsciences.github.io/2025/)

Opinion Maximization in Social Networks by Modifying Internal Opinions

通过修改内部意见实现社交网络中的意见最大化

Gengyu Wang, Runze Zhang, Zhongzhi Zhang

AI总结 本文提出通过修改关键节点内部意见来最大化社交网络整体意见，采用高效采样算法和确定性异步算法，实现高效率和精度的优化方案。

Comments Accepted by NeurIPS 2025

Journal ref NeurIPS 2025, Poster ID: 117421

Global Minimizers of Sigmoid Contrastive Loss

对Sigmoid对比损失的全局极小值

Kiril Bangachev, Guy Bresler, Iliyas Noman, Yury Polyanskiy

机构 * Department of Electrical Engineering and Computer Science（电气工程与计算机科学系） 纠错

AI总结 本文理论分析了Sigmoid对比损失中同步可训练逆温度和偏置的优势，并提出改进训练动态的重参数化方法。

Comments Author names listed in alphabetical order. NeurIPS 2025. New version includes some results on the geometry of CLIP in addition to geometry of SigLIP