大模型推理能力

FEval-TTC: Fair Evaluation Protocol for Test-Time Compute

Pavel Rumiantsev, Soumyasundar Pal, Yingxue Zhang, Mark Coates

机构 * McGill University（麦吉尔大学） ； Huawei Noah’s Ark Lab（华为诺亚实验室） 纠错

专题命中 测试时计算 ：test-time compute(title,abstract)；reasoning(abstract)；CoT(abstract)；分类 cs.CL、cs.LG

On the Bias of Next-Token Predictors Toward Systematically Inefficient Reasoning: A Shortest-Path Case Study

Riccardo Alberghi, Elizaveta Demyanenko, Luca Biggio, Luca Saglietti

专题命中 测试时计算 ：reasoning(title,abstract)；test-time compute(abstract)；分类 cs.CL、cs.AI、cs.LG

Generalizing Test-time Compute-optimal Scaling as an Optimizable Graph

Fali Wang, Jihai Chen, Shuhua Yang, Runxue Bao, Tianxiang Zhao, Zhiwei Zhang, Xianfeng Tang, Hui Liu, Qi He, Suhang Wang

机构 * The Pennsylvania State University（宾夕法尼亚州立大学） ； University of Pittsburgh（匹兹堡大学） ； Amazon（亚马逊） ； Microsoft（微软） 纠错

专题命中 测试时计算 ：test-time compute(title)；分类 cs.CL、cs.AI、cs.LG

Comments Under review

Representation Consistency for Accurate and Coherent LLM Answer Aggregation

Junqi Jiang, Tom Bewley, Salim I. Amoukou, Francesco Leofante, Antonio Rago, Saumitra Mishra, Francesca Toni

机构 * Imperial College London（帝国理工学院伦敦分校） ； J.P. Morgan AI Research（摩根大通人工智能研究） ； King’s College London（国王学院伦敦分校） 纠错

专题命中 测试时计算 ：reasoning(abstract)；分类 cs.CL、cs.LG

Comments Accepted at NeurIPS 2025. Camera-ready version

Continual Learning, Not Training: Online Adaptation For Agents

Aman Jaglan, Jarrod Barnes

机构 * Arc Intelligence 纠错

专题命中 测试时计算 ：reasoning(abstract)；分类 cs.AI、cs.LG

Comments 12 pages, 4 figures

GeoToken: Hierarchical Geolocalization of Images via Next Token Prediction

Narges Ghasemi, Amir Ziashahabi, Salman Avestimehr, Cyrus Shahabi

机构 * of Computer Science, University of Southern California, Los Angeles, CA, USA ； Computer Engineering, University of Southern California, Los Angeles, CA, USA 纠错

专题命中 测试时计算 ：test-time compute(abstract)；分类 cs.AI、cs.LG

Comments Accepted to IEEE International Conference on Data Mining (ICDM) 2025

TRISKELION-1: Unified Descriptive-Predictive-Generative AI

Nardeep Kumar, Arun Kanwar

机构 * Independent Machine Learning Researcher（独立研究者） 纠错

专题命中 测试时计算 ：reasoning(abstract)；分类 cs.AI、cs.LG

Comments 12 pages, 18 figures, submitted to arXiv (2025)

Cycle Consistency as Reward: Learning Image-Text Alignment without Human Preferences

Hyojin Bahng, Caroline Chan, Fredo Durand, Phillip Isola

机构 * MIT CSAIL（麻省理工学院计算机科学与人工智能实验室） 纠错

专题命中 测试时计算 ：verifier(abstract)；分类 cs.LG

LLM-Assisted Tool for Joint Generation of Formulas and Functions in Rule-Based Verification of Map Transformations

Ruidi He, Yu Zhang, Meng Zhang, Andreas Rausch

专题命中 测试时计算 ：verifier(abstract)

Contextual Relevance and Adaptive Sampling for LLM-Based Document Reranking

Jerry Huang, Siddarth Madala, Cheng Niu, Julia Hockenmaier, Tong Zhang

专题命中 测试时计算 ：reasoning(abstract)