语言大模型 / LLM - arXivDaily 专题

2510.03267 2026-02-02 cs.LG cs.AI 94%

PT$^2$-LLM: Post-Training Ternarization for Large Language Models

PT²-LLM：大语言模型的后训练三元化

Xianglong Yan, Chengzhu Bao, Zhiteng Li, Tianao Zhang, Kaicheng Yang, Haotong Qin, Ruobing Xie, Xingwu Sun, Yulun Zhang

机构 * Shanghai Jiao Tong University（上海交通大学）； ETH Zürich（苏黎世联邦理工学院）； Tencent Hunyuan（腾讯文言）

专题命中效率与部署：LLM(title,abstract)；large language model(title,abstract)；language model(title,abstract)；post-training(title,abstract)

AI总结 PT²-LLM通过后训练三元化技术，在降低内存成本的同时提升大语言模型的推理速度和效率。

Comments Accepted at ICLR 2026

2410.03129 2026-02-02 cs.CV cs.AI cs.CL cs.LG 93%

ARB-LLM: Alternating Refined Binarizations for Large Language Models

ARB-LLM: 交替精炼二值化用于大语言模型

Zhiteng Li, Xianglong Yan, Tianao Zhang, Haotong Qin, Dong Xie, Jiang Tian, zhongchao shi, Linghe Kong, Yulun Zhang, Xiaokang Yang

机构 * Shanghai Jiao Tong University（上海交通大学）； ETH Zürich（苏黎世联邦理工学院）； Lenovo Research（联想研究院）

专题命中效率与部署：LLM(title,abstract)；large language model(title,abstract)；language model(title,abstract)；post-training(abstract)

AI总结 ARB-LLM通过交替精炼二值化技术，有效解决LLM二值化中的分布偏移和列偏差问题，实现性能超越现有SOTA方法。

Comments The code and models will be available at https://github.com/ZHITENGLI/ARB-LLM

2601.22692 2026-02-02 cs.CL cs.AI cs.CR 90%

FNF: Functional Network Fingerprint for Large Language Models

FNF：大型语言模型的功能网络指纹

Yiheng Liu, Junhao Ning, Sichen Xia, Haiyang Sun, Yang Yang, Hanyang Chi, Xiaohui Gao, Ning Qiang, Bao Ge, Junwei Han, Xintao Hu

机构 * School of Automation, Northwestern Polytechnical University, Xi'an, China（自动化学院，西北工业大学，西安，中国）； School of Physics and Information Technology, Shaanxi Normal University, Xi'an, China（物理与信息技术学院，陕西师范大学，西安，中国）

专题命中效率与部署：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.AI

AI总结 FNF通过功能网络活动一致性检测LLM来源，提供非侵入性的知识产权保护方法。

Comments 13 pages, 4 figures

2601.22438 2026-02-02 cs.DC cs.CL cs.LG 90%

Towards Resiliency in Large Language Model Serving with KevlarFlow

面向大语言模型服务的韧性提升：KevlarFlow

Shangshu Qian, Kipling Liu, P. C. Sruthi, Lin Tan, Yongle Zhang

机构 * Department of XXX, University of YYY, Location, Country（XXX系，YYY大学，地点，国家）； School of ZZZ, Institute of WWW, Location, Country（ZZZ学院，WWW研究所，地点，国家）； Department of Computer Science, Purdue University, West Lafayette, IN, United States（计算机科学系，普渡大学，西拉法叶，印第安纳，美国）

专题命中效率与部署：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.CL、cs.LG

AI总结 KevlarFlow通过解耦模型并行初始化、动态流量重路由和背景KV缓存复制，显著提升了大语言模型服务的容错性和恢复效率。

2302.02662 2026-02-02 cs.LG 89%

Grounding Large Language Models in Interactive Environments with Online Reinforcement Learning

通过在线强化学习将大语言模型接地于交互环境

Thomas Carta, Clément Romac, Thomas Wolf, Sylvain Lamprier, Olivier Sigaud, Pierre-Yves Oudeyer

机构 * Inria (Flowers)（Inria（Flowers））； University of Bordeaux（波尔多大学）； Hugging Face ； Univ Angers, LERIA, SFR MATHSTIC（昂热大学，LERIA，SFR MATHSTIC）； Sorbonne Université（索邦大学）

专题命中效率与部署：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.LG

AI总结本文提出GLAM方法，通过在线强化学习将大语言模型接地于交互环境，以提升样本效率和泛化能力，并探讨在线学习的影响。

Comments The associated code can be found at https://github.com/flowersteam/Grounding_LLMs_with_online_RL. This is an extended version of the paper published at ICML 2023: https://proceedings.mlr.press/v202/carta23a

Journal ref PMLR 202 (2023):3676-3713

2601.22779 2026-02-02 eess.AS cs.SD 89%

Streaming Speech Recognition with Decoder-Only Large Language Models and Latency Optimization

流式语音识别与解码器-only大型语言模型及延迟优化

Genshun Wan, Wenhui Zhang, Jing-Xuan Zhang, Shifu Xiong, Jianqing Gao, Zhongfu Ye

机构 * University of Science and Technology of China（科学技术大学）； iFLYTEK Research, iFLYTEK Co., Ltd.（iFLYTEK研究院，iFLYTEK有限公司）； School of Artificial Intelligence and Computer Science（人工智能与计算机科学学院）

专题命中效率与部署：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)

AI总结本文提出了一种结合解码器-only大型语言模型和延迟优化的流式语音识别方法，通过动态分割语音嵌入和联合训练策略，提升了识别性能并降低了延迟。

Comments accepted to ICASSP 2026

2505.12260 2026-02-02 cs.IR cs.AI cs.CL 86%

LightRetriever: A LLM-based Text Retrieval Architecture with Extremely Faster Query Inference

LightRetriever: 一种基于大语言模型的文本检索架构，具有极快的查询推理速度

Guangyuan Ma, Yongliang Ma, Xuanrui Gou, Zhenpeng Su, Ming Zhou, Songlin Hu

机构 * Institute of Information Engineering, Chinese Academy of Sciences（中国科学院信息工程研究所）； School of Cyber Security, University of Chinese Academy of Sciences（中国科学院大学网络安全学院）； Langboat Technology（兰舟科技）

专题命中效率与部署：LLM(title,abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI

AI总结 LightRetriever通过轻量级查询编码器显著提升检索效率，实现查询推理速度提升超1000倍，端到端吞吐量提升超10倍，且在多种任务中保持95%的检索性能。

Comments Accepted by ICLR 2026

2510.26374 2026-02-02 cs.AI 85%

BOTS: A Unified Framework for Bayesian Online Task Selection in LLM Reinforcement Finetuning

BOTS：一种用于LLM强化微调中贝叶斯在线任务选择的统一框架

Qianli Shen, Daoyuan Chen, Yilun Huang, Zhenqing Ling, Yaliang Li, Bolin Ding, Jingren Zhou

机构 * Alibaba Group（阿里巴巴集团）

专题命中效率与部署：LLM(title,abstract)；large language model(abstract)；language model(abstract)；分类 cs.AI

AI总结 BOTS通过贝叶斯推断和汤普森采样，提高LLM强化微调中任务选择的效率和性能。

Comments Accepted as a conference paper at ICLR 2026

2510.06213 2026-02-02 cs.LG 85%

Training Dynamics Impact Post-Training Quantization Robustness

训练动态对训练后量化鲁棒性的影响

Albert Catalan-Tatjer, Niccolò Ajroldi, Jonas Geiping

机构 * ELLIS Institute Tübingen（图宾根ELLIS研究所）； Max Planck Institute for Intelligent Systems（智能系统马克斯·普朗克研究所）； Tübingen AI Center（图宾根人工智能中心）

专题命中效率与部署：post-training(title,abstract)；large language model(abstract)；language model(abstract)；分类 cs.LG

AI总结研究发现训练动态中的学习率衰减会影响量化鲁棒性，通过调整训练超参数可提升大规模量化效果。

2601.22362 2026-02-02 cs.LG 85%

Understanding Efficiency: Quantization, Batching, and Serving Strategies in LLM Energy Use

理解效率：在LLM能耗中的量化、批量和服务策略

Julien Delavande, Regis Pierrard, Sasha Luccioni

机构 * Hugging Face

专题命中效率与部署：LLM(title,abstract)；large language model(abstract)；language model(abstract)；分类 cs.LG

AI总结本文研究了LLM推理能耗中的量化、批量和服务策略，发现低精度和批量处理可显著降低能耗，同时强调服务堆栈的协调对可持续部署的重要性。

2601.22318 2026-02-02 cs.LG 85%

Federate the Router: Learning Language Model Routers with Sparse and Decentralized Evaluations

联邦路由器：利用稀疏和去中心化的评估学习语言模型路由器

Baris Askin, Shivam Patel, Anupam Nayak, Andrea Vigano, Jiin Woo, Gauri Joshi, Carlee Joe-Wong

机构 * Carnegie Mellon University（卡内基梅隆大学）

专题命中效率与部署：language model(title,abstract)；LLM(abstract)；large language model(abstract)；分类 cs.LG

AI总结本文提出首个联邦框架，通过稀疏和去中心化评估学习语言模型路由器，提升查询路由的准确性和成本效率。

2601.22996 2026-02-02 cs.DS 85%

Competitive Non-Clairvoyant KV-Cache Scheduling for LLM Inference

竞争性非clairvoyant KV缓存调度用于LLM推理

Yiding Feng, Zonghan Yang, Yuhao Zhang

专题命中效率与部署：LLM(title,abstract)；large language model(abstract)；language model(abstract)

AI总结本文提出非clairvoyant的几何切片算法（GSA）和clairvoyant的几何批处理算法（GBA），在LLM推理中实现首个常数竞争比，显著提升性能界。

2601.22791 2026-02-02 cs.SE 85%

Understanding on the Edge: LLM-generated Boundary Test Explanations

对边界的理解：LLM生成的边界测试解释

Sabinakhon Akbarova, Felix Dobslaw, Robert Feldt

专题命中效率与部署：LLM(title,abstract)；large language model(abstract)；language model(abstract)

AI总结本文研究LLM生成的边界解释在软件测试中的有效性，通过调查和访谈发现，清晰结构、权威来源引用和可操作示例能提升解释质量，提出七项要求的检查表以指导未来工具设计。

Comments This is the author's accepted manuscript of a paper accepted for publication at AST 2026. The final version will appear in the ACM Digital Library

2601.18220 2026-02-02 cs.SD eess.AS 85%

LLM-ForcedAligner: A Non-Autoregressive and Accurate LLM-Based Forced Aligner for Multilingual and Long-Form Speech

LLM-ForcedAligner: 一种非自回归且准确的基于大语言模型的强制对齐器，适用于多语言和长文本语音

Bingshen Mu, Xian Shi, Xiong Wang, Hexin Liu, Jin Xu, Lei Xie

机构 * College of Computing and Data Science, Nanyang Technological University, Singapore（计算与数据科学学院，南洋理工大学，新加坡）

专题命中效率与部署：LLM(title,abstract)；large language model(abstract)；language model(abstract)

AI总结 LLM-ForcedAligner通过非自回归方法实现多语言和长文本语音的准确强制对齐，显著减少时间偏移。

2601.23000 2026-02-02 cs.LG cs.AI 84%

Mano: Restriking Manifold Optimization for LLM Training

Mano：重新审视面向大语言模型训练的流形优化

Yufei Gu, Zeke Xie

机构 * xLeaF Lab, The Hong Kong University of Science and Technology (Guangzhou)（xLeaF实验室，香港科学与技术大学（广州））

专题命中效率与部署：LLM(title)；large language model(abstract)；language model(abstract)；分类 cs.AI、cs.LG

AI总结 Mano是一种新颖高效的流形优化器，通过创新投影和约束方法，在训练大语言模型时显著优于AdamW和Muon。

2601.17687 2026-02-02 cs.LG cs.AI 84%

Agentic reinforcement learning empowers next-generation chemical language models for molecular design and synthesis

代理强化学习赋能下一代化学语言模型用于分子设计与合成

Hao Li, He Cao, Shenyao Peng, Zijing Liu, Bin Feng, Yu Wang, Zhiyuan Yan, Yonghong Tian, Yu Li, Li Yuan

机构 * School of Electronic and Computer Engineering, Peking University（电子与计算机工程学院，北京大学）； Shenzhen Graduate School, Shenzhen, 518055, China（深圳研究生院，深圳，518055，中国）； International Digital Economy Academy (IDEA)（国际数字经济学院（IDEA））； Peng Cheng Laboratory, Shenzhen, 518000, China（鹏城实验室，深圳，518000，中国）

专题命中效率与部署：language model(title,abstract)；small language model(abstract)；分类 cs.AI、cs.LG

AI总结 ChemCRAFT通过代理强化学习实现化学语言模型的高效分子设计与合成，突破传统模型在成本与隐私间的限制。

Comments Working in Progress, 13 pages, 4 figures

2504.14366 2026-02-02 cs.CL cs.AI 84%

What Matters in Linearizing Language Models? A Comparative Study of Architecture, Scale, and Task Adaptation

语言模型线性化中什么更重要？架构、规模和任务适应的比较研究

Patrick Haller, Jonas Golde, Alan Akbik

机构 * Humboldt-Universität zu Berlin（柏林洪堡大学）

专题命中效率与部署：language model(title,abstract)；instruction tuning(abstract)；分类 cs.CL、cs.AI

AI总结本研究比较了七种架构在线性化语言模型中的表现，发现架构归纳偏置是影响线性化成功的关键因素，而非单纯扩大训练计算量。

Comments 9 pages

2601.23253 2026-02-02 cs.CV cs.LG 83%

Training-Free Test-Time Adaptation with Brownian Distance Covariance in Vision-Language Models

无需训练的测试时适应与布朗距离协方差在视觉-语言模型中

Yi Zhang, Chun-Wun Cheng, Angelica I. Aviles-Rivero, Zhihai He, Liang-Jie Zhang

机构 * College of Computer Science and Software Engineering, Shenzhen University, China（深圳大学计算机科学与软件工程学院）； University of Cambridge, Cambridge, UK（剑桥大学）； Yau Mathematical Sciences Center, Tsinghua University, Beijing, China（清华大学尤里伊数学科学中心）； Southern University of Science and Technology, Shenzhen, China（南方科技大学）

专题命中效率与部署：language model(title,abstract)；prompting(abstract)；分类 cs.LG

AI总结 TaTa通过布朗距离协方差实现无需训练的测试时适应，提升视觉-语言模型在领域偏移下的效率与稳定性，同时在泛化性能上取得突破。

Comments Accepted in ICASSP 2026

2508.06309 2026-02-02 cs.CL math.PR 83%

Matrix-Driven Identification and Reconstruction of LLM Weight Homology

基于矩阵的LLM权重同源识别与重建

Ruichong Zhang, Daniel Goldstein

机构 * Qiuzhen College, Tsinghua University, Beijing, China（清华大学邱臻学院）

专题命中效率与部署：LLM(title)；large language model(abstract)；language model(abstract)；分类 cs.CL

AI总结 MDIR是一种通过矩阵分析实现大语言模型权重同源识别与重建的先进方法，无需模型推理即可准确检测权重对应关系并评估其统计显著性。

2505.12767 2026-02-02 cs.AI 83%

Language Models That Walk the Talk: A Framework for Formal Fairness Certificates

语言模型言出必行：一个形式公平证书的框架

Danqing Chen, Tobias Ladner, Ahmed Rayen Mhadhbi, Matthias Althoff

机构 * Technical University of Munich, Germany（慕尼黑技术大学）

专题命中效率与部署：language model(title,abstract)；large language model(abstract)；分类 cs.AI

AI总结本文提出一个框架，用于验证大语言模型的鲁棒性和公平性，特别是在性别公平和毒性检测中的应用。

2601.23281 2026-02-02 cs.CV 82%

User Prompting Strategies and Prompt Enhancement Methods for Open-Set Object Detection in XR Environments

面向XR环境的开放集目标检测中的用户提示策略与提示增强方法

Junfeng Lin, Yanming Xiu, Maria Gorlatova

机构 * Department of Electrical and Computer Engineering, Duke University（电子工程系，杜克大学）

专题命中效率与部署：prompting(title,abstract)；language model(abstract)

AI总结本文提出针对XR环境中的开放集目标检测，通过研究用户提示策略和增强方法，提升模型在模糊提示下的鲁棒性，实验结果显示提示增强可使mIoU提升超过55%。

Comments Accepted by IEEE VR 2026: GenAI-XR workshop

2506.06446 2026-02-02 cs.CL cs.AI cs.LG stat.ML 82%

Tokenization Multiplicity Leads to Arbitrary Price Variation in LLM-as-a-service

令牌化多样性导致LLM-as-a-service中的任意价格波动

Ivi Chatzi, Nina Corvelo Benz, Stratis Tsirtsis, Manuel Gomez-Rodriguez

专题命中效率与部署：LLM(title,abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结本文提出规范生成方法，通过限制LLM生成唯一令牌化以解决LLM-as-a-service中令牌化多样性导致的价格波动问题。

2601.17768 2026-02-02 cs.LG cs.AI cs.DC 82%

LLM-42: Enabling Determinism in LLM Inference with Verified Speculation

LLM-42: 通过验证推测实现LLM推理的确定性

Raja Gond, Aditya K Kamath, Ramachandran Ramjee, Ashish Panwar

机构 * Microsoft Research（微软研究院）； University of Washington（华盛顿大学）

专题命中效率与部署：LLM(title,abstract)；分类 cs.AI、cs.LG

AI总结 LLM-42通过验证推测方法，在LLM推理中实现确定性，利用形状一致的减少和轻量级验证-回滚循环，减少确定性需求的开销。

Comments https://github.com/microsoft/llm-42

2601.22657 2026-02-02 cs.CL cs.AI 81%

NAG: A Unified Native Architecture for Encoder-free Text-Graph Modeling in Language Models

NAG：一种用于语言模型中无编码文本图建模的统一原生架构

Haisong Gong, Zhibo Liu, Qiang Liu, Shu Wu, Liang Wang

机构 * New Laboratory of Pattern Recognition (NLPR), Institute of Automation, Chinese Academy of Sciences, Beijing, China ； School of Artificial Intelligence, University of Chinese Academy of Sciences (UCAS), Beijing, China ； School of Advanced Interdisciplinary Sciences, UCAS, Beijing, China

专题命中效率与部署：language model(title,abstract)；分类 cs.CL、cs.AI

AI总结 NAG提出一种统一的原生架构，将图处理内化于语言模型，通过自注意力机制和位置ID校准实现结构依赖和等价性，提升文本图建模的连贯性和效率。

2601.22752 2026-02-02 cs.LG 79%

OSNIP: Breaking the Privacy-Utility-Efficiency Trilemma in LLM Inference via Obfuscated Semantic Null Space

OSNIP: 通过混淆语义空域打破LLM推理中的隐私-效用-效率三元困境

Zhiyuan Cao, Zeyu Ma, Chenhao Yang, Han Zheng, Mingang Chen

机构 * Shanghai Key Laboratory of Computer Software Testing and Evaluating（上海计算机软件测试与评估重点实验室）； Shanghai Normal University（上海师范大学）； Shanghai Polytechnic University（上海理工大学）； TrustAI Pte. Ltd.（TrustAI有限公司）

专题命中效率与部署：LLM(title,abstract)；分类 cs.LG

AI总结 OSNIP通过混淆语义空域技术，在LLM推理中实现隐私、效用和效率的平衡，有效降低攻击成功率并保持模型性能。

2601.22688 2026-02-02 cs.CL 79%

TSLM: Tree-Structured Language Modeling for Divergent Thinking

TSLM：用于发散思维的树状语言建模

Doyoung Kim, Jaehyeok Doo, Minjoon Seo

机构 * NYU（纽约大学）； KAIST（韩国科学技术院）

专题命中效率与部署：language model(title,abstract)；分类 cs.CL

AI总结 TSLM通过树状结构语言建模实现发散思维，利用完整搜索树训练提升推理效率和鲁棒性。

2601.22371 2026-02-02 cs.LG 79%

FIRE: Multi-fidelity Regression with Distribution-conditioned In-context Learning using Tabular Foundation Models

FIRE: 基于分布条件化上下文学习的多保真度回归

Rosen Ting-Ying Yu, Nicholas Sung, Faez Ahmed

机构 * Department of Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, MA, USA（麻省理工学院机械工程系）； Center for Computational Engineering, Massachusetts Institute of Technology, Cambridge, MA, USA（麻省理工学院计算工程中心）

专题命中效率与部署：foundation model(title,abstract)；分类 cs.LG

AI总结 FIRE通过分布条件化上下文学习，实现高效的多保真度回归，提升准确性和不确定性量化，优于现有方法。

2505.22654 2026-02-02 cs.CV cs.CL 79%

VScan: Rethinking Visual Token Reduction for Efficient Large Vision-Language Models

VScan：重新思考视觉标记减少以提高高效的大视觉-语言模型

Ce Zhang, Kaixin Ma, Tianqing Fang, Wenhao Yu, Hongming Zhang, Zhisong Zhang, Haitao Mi, Dong Yu

机构 * Carnegie Mellon University（卡内基梅隆大学）； Tencent AI Lab（腾讯AI实验室）

专题命中效率与部署：language model(title,abstract)；分类 cs.CL

AI总结 VScan通过两阶段视觉标记减少框架提升大视觉-语言模型的推理效率，实现2.91倍预填充加速和10倍FLOPs减少，性能损失仅0.6%

Comments Accepted at TMLR 2026. Project page: https://zhangce01.github.io/VScan/

2601.22959 2026-02-02 cs.CV 78%

Triage: Hierarchical Visual Budgeting for Efficient Video Reasoning in Vision-Language Models

Triage: 分层视觉预算机制用于视觉语言模型中的高效视频推理

Anmin Wang, Nan Zhang, Wei Tao, Xiaoyang Qu, Guokuan Li, Jiguang Wan, Jianzong Wang

机构 * Huazhong University of Science and Technology（华中科技大学）； Ping An Technology (Shenzhen) Co., Ltd.（平安科技（深圳）有限公司）

专题命中效率与部署：language model(title,abstract)

AI总结 Triage通过分层视觉预算机制优化视频推理，提升效率并保持性能。

Comments Accepted to 2026 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2026)

2601.22738 2026-02-02 cs.CV 78%

StreamSense: Streaming Social Task Detection with Selective Vision-Language Model Routing

StreamSense: 基于选择性视觉-语言模型路由的流式社交任务检测

Han Wang, Deyi Ji, Lanyun Zhu, Jiebo Luo, Roy Ka-Wei Lee

机构 * Singapore University of Technology and Design（新加坡科技设计大学）； University of Science and Technology of China（中国科学技术大学）； Nanyang Technological University（南洋理工大学）； University of Rochester（罗切斯特大学）

专题命中效率与部署：language model(title,abstract)

AI总结 StreamSense通过轻量级编码器与选择性路由结合VLM专家，提升流式社交任务检测的准确性和效率。

Comments 10 pages, 4 figures, The Web Conference 2026

1. 效率与部署 65 篇

PT$^2$-LLM: Post-Training Ternarization for Large Language Models

ARB-LLM: Alternating Refined Binarizations for Large Language Models

FNF: Functional Network Fingerprint for Large Language Models

Towards Resiliency in Large Language Model Serving with KevlarFlow

Grounding Large Language Models in Interactive Environments with Online Reinforcement Learning

Streaming Speech Recognition with Decoder-Only Large Language Models and Latency Optimization

LightRetriever: A LLM-based Text Retrieval Architecture with Extremely Faster Query Inference

BOTS: A Unified Framework for Bayesian Online Task Selection in LLM Reinforcement Finetuning

Training Dynamics Impact Post-Training Quantization Robustness

Understanding Efficiency: Quantization, Batching, and Serving Strategies in LLM Energy Use

Federate the Router: Learning Language Model Routers with Sparse and Decentralized Evaluations

Competitive Non-Clairvoyant KV-Cache Scheduling for LLM Inference

Understanding on the Edge: LLM-generated Boundary Test Explanations

LLM-ForcedAligner: A Non-Autoregressive and Accurate LLM-Based Forced Aligner for Multilingual and Long-Form Speech

Mano: Restriking Manifold Optimization for LLM Training

Agentic reinforcement learning empowers next-generation chemical language models for molecular design and synthesis

What Matters in Linearizing Language Models? A Comparative Study of Architecture, Scale, and Task Adaptation

Training-Free Test-Time Adaptation with Brownian Distance Covariance in Vision-Language Models

Matrix-Driven Identification and Reconstruction of LLM Weight Homology

Language Models That Walk the Talk: A Framework for Formal Fairness Certificates

User Prompting Strategies and Prompt Enhancement Methods for Open-Set Object Detection in XR Environments

Tokenization Multiplicity Leads to Arbitrary Price Variation in LLM-as-a-service

LLM-42: Enabling Determinism in LLM Inference with Verified Speculation

NAG: A Unified Native Architecture for Encoder-free Text-Graph Modeling in Language Models

OSNIP: Breaking the Privacy-Utility-Efficiency Trilemma in LLM Inference via Obfuscated Semantic Null Space

TSLM: Tree-Structured Language Modeling for Divergent Thinking

FIRE: Multi-fidelity Regression with Distribution-conditioned In-context Learning using Tabular Foundation Models

VScan: Rethinking Visual Token Reduction for Efficient Large Vision-Language Models

Triage: Hierarchical Visual Budgeting for Efficient Video Reasoning in Vision-Language Models

StreamSense: Streaming Social Task Detection with Selective Vision-Language Model Routing