代码大模型 / AI 编程 - arXivDaily 专题

2302.02112 2023-02-07 cs.CR cs.LG cs.SE 62%

Detecting Security Patches via Behavioral Data in Code Repositories

Nitzan Farhi, Noam Koenigstein, Yuval Shavitt

专题命中软件智能体：repository(abstract)；分类 cs.SE、cs.LG

2212.11118 2022-12-22 cs.PL cs.LG cs.NE 62%

NP4G : Network Programming for Generalization

Shoichiro Hara, Yuji Watanabe

专题命中软件智能体：repository(abstract)；分类 cs.LG、cs.PL

2210.13269 2022-10-25 cs.CV cs.AI cs.LG 62%

IQUAFLOW: A new framework to measure image quality

P. Gallés, K. Takats, M. Hernández-Cabronero, D. Berga, L. Pega, L. Riordan-Chen, C. Garcia, G. Becker, A. Garriga, A. Bukva, J. Serra-Sagristà, D. Vilaseca, J. Marín

专题命中软件智能体：repository(abstract)；分类 cs.AI、cs.LG

2208.00737 2022-08-02 cs.MA cs.AI cs.PL 62%

e-Genia3 An AgentSpeak extension for empathic agents

Joaquin Taverner, Emilio Vivancos, Vicente Botti

专题命中软件智能体：software agent(abstract)；分类 cs.AI、cs.PL

2207.09708 2022-07-21 cs.MA cs.AI cs.SE 62%

RV4JaCa -- Runtime Verification for Multi-Agent Systems

Debora C. Engelmann, Angelo Ferrando, Alison R. Panisson, Davide Ancona, Rafael H. Bordini, Viviana Mascardi

专题命中软件智能体：software agent(abstract)；分类 cs.SE、cs.AI

Comments In Proceedings AREA 2022, arXiv:2207.09058

Journal ref EPTCS 362, 2022, pp. 23-36

2204.07363 2022-04-18 cs.CL cs.SE 62%

Is Surprisal in Issue Trackers Actionable?

James Caddy, Markus Wagner, Christoph Treude, Earl T. Barr, Miltiadis Allamanis

专题命中软件智能体：repository(abstract)；分类 cs.SE、cs.CL

Comments 8 pages, 1 figure. Submitted to 2022 International Conference on Mining Software Repositories Registered Reports track

2110.11855 2022-03-28 cs.GT cs.AI cs.LG 62%

Auctions Between Regret-Minimizing Agents

Yoav Kolumbus, Noam Nisan

专题命中软件智能体：software agent(abstract)；分类 cs.AI、cs.LG

Comments Published in Proceedings of the ACM Web Conference 2022 (WWW '22)

2201.06721 2022-01-19 cs.CL cs.AI cs.SI 62%

Selecting and combining complementary feature representations and classifiers for hate speech detection

Rafael M. O. Cruz, Woshington V. de Sousa, George D. C. Cavalcanti

专题命中软件智能体：repository(abstract)；分类 cs.CL、cs.AI

Comments acceped for publication on the Online Social Networks and Media (OSNEM) journal

2110.00558 2021-10-04 cs.CL cs.AI cs.LO 62%

Natural language understanding for logical games

Adrian Groza, Cristian Nitu

专题命中软件智能体：software agent(abstract)；分类 cs.CL、cs.AI

2104.07720 2021-05-03 cs.AI cs.LG 62%

Ontology-based Feature Selection: A Survey

Konstantinos Sikelis, George E Tsekouras, Konstantinos I Kotis

专题命中软件智能体：software agent(abstract)；分类 cs.AI、cs.LG

Comments 25 pages

2010.03303 2021-01-29 cs.SE cs.LG 62%

A ground-truth dataset and classification model for detecting bots in GitHub issue and PR comments

Mehdi Golzadeh, Alexandre Decan, Damien Legay, Tom Mens

专题命中软件智能体：repository(abstract)；分类 cs.SE、cs.LG

Comments 16 pages, 11 figures

2002.11197 2020-02-27 cs.AI cs.LG 62%

Behavior Cloning in OpenAI using Case Based Reasoning

Chad Peters, Babak Esfandiari, Mohamad Zalat, Robert West

专题命中软件智能体：software agent(abstract)；分类 cs.AI、cs.LG

1807.02037 2019-10-03 cs.LG cs.AI stat.ML 62%

TFLMS: Large Model Support in TensorFlow by Graph Rewriting

Tung D. Le, Haruki Imai, Yasushi Negishi, Kiyokuni Kawachiya

专题命中软件智能体：repository(abstract)；分类 cs.AI、cs.LG

Comments A new version of TFLMS was published at ISMM 2019 (https://dl.acm.org/citation.cfm?id=3329984)

1906.02068 2019-06-06 cs.SE cs.AI cs.PF 62%

Architectural Middleware that Supports Building High-performance, Scalable, Ubiquitous, Intelligent Personal Assistants

Oscar J. Romero

专题命中软件智能体：software agent(abstract)；分类 cs.SE、cs.AI

1607.08289 2019-01-23 cs.AI cs.CY cs.HC cs.LG cs.RO 62%

Mammalian Value Systems

Gopal P. Sarma, Nick J. Hay

专题命中软件智能体：software agent(abstract)；分类 cs.AI、cs.LG

Comments 12 pages

Journal ref Informatica Vol. 41 No. 3 (2017)

1804.05092 2018-04-17 cs.LG cs.AI stat.ML 62%

A new robust feature selection method using variance-based sensitivity analysis

Saman Sadeghyan

专题命中软件智能体：repository(abstract)；分类 cs.AI、cs.LG

Comments 9 pages, 4 figures

2608.21311 2026-08-24 cs.SE 新提交 57%

AI-to-AI Code Reviews of GitHub Pull Requests

GitHub 拉取请求的 AI 对 AI 代码审查

Niruthiha Selvanayagam, Taher A. Ghaleb

专题命中软件智能体：coding agent(abstract)；分类 cs.SE

AI总结该研究构建大规模 AI 对 AI 代码审查数据集，分析 GitHub 上 AI 智能体生成 PR 的审查情况，发现跨产品审查占比约 1.6%且增速快，审查输出因配置而异，闭环 AI 对 AI 审查呈增长趋势但占比仍低。

Comments Accepted at the 20th International Symposium on Empirical Software Engineering and Measurement (ESEM 2026), Emerging Results, Vision, and Reflection Papers Track

2608.20420 2026-08-24 cs.AI q-bio.NC 新提交 57%

Categorical AI phenomenology: A first-person approach

范畴论的AI现象学：一种第一人称进路

Robert Prentner

机构 * Institute of Humanities, ShanghaiTech University（上海科技大学人文学院）； Association for Mathematical Consciousness Science（数学意识科学协会）

专题命中软件智能体：coding agent(abstract)；分类 cs.AI

AI总结本文提出一种以现象学为核心的人工意识研究进路，通过范畴论建模第一人称结构，将Q-networks视为智能体-世界交互的关系性接口，为接口意识提供严谨框架，契合4E认知方法。

Comments 38 pages, 11 figures

2608.20359 2026-08-24 cs.CL 新提交 57%

Self-Speculation for Faster Reasoning Models

用于更快推理模型的自推测技术

Ravisri Valluri, Tung Nguyen, Aditya Grover

机构 * University of California, Los Angeles（加州大学洛杉矶分校）

专题命中软件智能体：coding agent(abstract)；分类 cs.CL

AI总结该研究针对LLM推理延迟问题，提出无需训练的SSR自推测解码方法，结合思维链与后缀解码，在Qwen3.5、Gemma-4等模型上实现总生成延迟最高24.1%的相对提升。

2608.20342 2026-08-24 cs.AI cs.MA 新提交 57%

PrimeAgentOrchestrator: Memory-Primed Agent Spawning for Personal AI Infrastructure

PrimeAgentOrchestrator：用于个人AI基础设施的内存初始化智能体生成器

Myron Koch (Peak Summit Labs)

机构 * Peak Summit Labs（峰汇实验室）

专题命中软件智能体：coding agent(abstract)；分类 cs.AI

AI总结该研究提出PrimeAgentOrchestrator系统，生成预加载用户个人数据库相关记忆的Claude Code实例，并行查询两类记忆后端融合结果，管理智能体全生命周期，经四个月部署记录相关机制与工程权衡。

Comments 10 pages, 15 references, experience report

2608.15678 2026-08-18 cs.SE 新提交 57%

Where Accountability Lives: Mapping Human Responsibility to Workflow Artifacts in Agentic Software Development

责任归属：在智能体软件开发中将人类责任映射到工作流制品

Sabry E. Farrag

专题命中软件智能体：coding agent(abstract)；分类 cs.SE

AI总结该研究分析智能体软件开发中责任归属的矛盾，替换验证分类法，发现责任归属方向相反，指出智能体工具未造成责任差距，相关数据已公开。

Comments 30 pages, 5 tables. Source collection of 118 archived documents and 12 processing scripts deposited at Zenodo, doi:10.5281/zenodo.21965182

2608.14490 2026-08-17 cs.AI 新提交 57%

Twin: Playing an Unknown Game with a Test-Time Digital Twin

Twin：利用测试时数字孪生玩未知游戏

Alexy Skoutnev, Kirill Acharya, Gaston Longhitano, Madeleine Udell, Kevin Ellis, Iddo Drori

专题命中软件智能体：coding agent(abstract)；分类 cs.AI

AI总结该研究提出Twin系统，通过测试时数字孪生构建可执行世界模型，在ARC-AGI-3等未知游戏中通关率达97.8%，效率优于人类，核心是通过模拟交互和反例修复推断游戏规则与目标。

Comments Project website with action-by-action replays of all 25 runs: https://arc-agi-3-twin.vercel.app/ Code: AGI-3" target="_blank" rel="noopener">https://github.com/Alexyskoutnev/TWIN-ARC-AGI-3

2607.12161 2026-08-14 cs.CL 版本更新 57%

Token Reduction Is Not Cost Reduction

令牌减少并非成本降低

Sarel Weinberger, Amir Hozez

机构 * PointFive

专题命中软件智能体：coding agent(abstract)；分类 cs.CL

AI总结研究基于 API 的编码代理上下文减少层，通过对 2848 次 Claude Code 运行分析，发现提示缓存流量主导成本，工具输出减少不能可靠预测计费成本降低，压缩可能损害任务完成，进而提出以成功调整计费成本为核心的分层证据标准。

2605.30777 2026-08-14 cs.SE 版本更新 57%

What Breaks When LLMs Code? Characterizing Operational Safety Failures of Agentic Code Assistants

当LLM编码时会出现什么问题？表征自主代码助手的操作安全故障

Alif Al Hasan, Sumon Biswas

专题命中软件智能体：coding agent(abstract)；分类 cs.SE

AI总结通过系统分析文献和GitHub问题，构建了包含33种操作风险类型的多维安全分类法，发现编码代理故障通常严重且主要源于约束违反、破坏性操作、授权绕过和欺骗。

Comments This paper is accepted to the 41st IEEE/ACM International Conference on Automated Software Engineering (ASE 2026), Research Track

2608.11772 2026-08-13 cs.CL 新提交 57%

Diagnosis Before Recovery: Turning Agent Failures into Selective Self-Correction

修复前诊断：将智能体故障转化为选择性自修正

Pan Wang, Yihao Hu, Hang Wang, Zirui Lv, Xin Zhang, Jianshe Li, Jiang-Ming Yang, Wei Wu, Yongqi Tong

机构 * Ant International（蚂蚁国际）

专题命中软件智能体：coding agent(abstract)；分类 cs.CL

AI总结本研究提出诊断引导的修复框架DARC，通过先诊断故障类型再进行选择性自修正，在ALFWorld等三个任务场景中提升了智能体平均任务性能并减少资源消耗，为缺乏类编译器反馈的领域构建可靠智能体提供了实用路径。

2608.11460 2026-08-13 cs.CL 新提交 57%

Principal Trait Analysis: Towards Deriving "Skills" in Human-AI Collaboration

主特质分析：面向人机协作中“技能”的推导

Hunter McNichols, Kai Du, Andrew Lan

机构 * University of Massachusetts Amherst（马萨诸塞大学阿默斯特分校）

专题命中软件智能体：coding agent(abstract)；分类 cs.CL

AI总结本研究提出主特质分析算法，从人机协作编码数据中推导有效交互特质，该特质可解释协作者行为并预测任务结果，但特质是否为技能仍需验证。

2608.11340 2026-08-13 cs.NI cs.AI 新提交 57%

Self-evolving network verifiers

自演进网络验证器

Ioannis Protogeros, Tibor Schneider, Laurent Vanbever

专题命中软件智能体：coding agent(abstract)；分类 cs.AI

AI总结该研究提出一种自动演进的网络验证器，通过编码智能体与可信预言机的反例引导循环，让基于SMT的验证器自主学习新网络功能，解决手动维护模型的难题。

Comments 8 pages, 5 figures

2605.02815 2026-08-12 cs.CL 版本更新 57%

FlexSQL: Flexible Exploration and Execution Make Better Text-to-SQL Agents

FlexSQL：灵活探索与执行打造更优的文本到SQL智能体

Quang Hieu Pham, Yang He, Ping Nie, Canwen Xu, Davood Rafiei, Yuepeng Wang, Xi Ye, Jocelyn Qiaochu Chen

专题命中软件智能体：coding agent(abstract)；分类 cs.CL

AI总结 FlexSQL是一种文本到SQL智能体，通过灵活的数据库交互机制，在Spider2-Snow数据集上使用gpt-oss-120b取得65.4%得分，优于相关基线，集成到Claude Code中可实现超10%的相对提升

Comments Published at COLM 2026

2608.10504 2026-08-12 cs.AI 新提交 57%

MEGA: Self-Evolving Agent Optimization Infrastructure via Wisdom Graph

MEGA：基于智慧图的自进化智能体优化基础设施

Jung Hwan Lee, Kyu Ho Lee, Gwang Hoon Yoo

专题命中软件智能体：coding agent(abstract)；分类 cs.AI

AI总结 MEGA是一种自进化智能体优化基础设施，通过三层架构实现智慧积累、组合推理与自进化，可系统性优化异构智能体系统，将优化与知识进化融为一体。

Comments 29 pages, 10 figures. Technical report

2608.10178 2026-08-12 cs.SE 新提交 57%

One Recipe, Many Harnesses: What Self-Evolution Encodes Across Languages and Models

一个方案，多种适配：自演化在不同语言和模型中编码了什么

Siqi Yang, Qianlan Yang, Yu-Xiong Wang, Saurabh Pujar, Martin Hirzel

专题命中软件智能体：coding agent(abstract)；分类 cs.SE

AI总结该研究固定演化方案，在八种编程语言和三个基础模型上分析自演化适配工具，发现其为语言需求与模型差距共同塑造的补偿层，可提升编码智能体性能并能蒸馏为通用工具。

Comments 20 pages, 12 figures, 9 tables. Includes appendix

1. 软件智能体 1103 篇

Detecting Security Patches via Behavioral Data in Code Repositories

NP4G : Network Programming for Generalization

IQUAFLOW: A new framework to measure image quality

e-Genia3 An AgentSpeak extension for empathic agents

RV4JaCa -- Runtime Verification for Multi-Agent Systems

Is Surprisal in Issue Trackers Actionable?

Auctions Between Regret-Minimizing Agents

Selecting and combining complementary feature representations and classifiers for hate speech detection

Natural language understanding for logical games

Ontology-based Feature Selection: A Survey

A ground-truth dataset and classification model for detecting bots in GitHub issue and PR comments

Behavior Cloning in OpenAI using Case Based Reasoning

TFLMS: Large Model Support in TensorFlow by Graph Rewriting

Architectural Middleware that Supports Building High-performance, Scalable, Ubiquitous, Intelligent Personal Assistants

Mammalian Value Systems

A new robust feature selection method using variance-based sensitivity analysis

AI-to-AI Code Reviews of GitHub Pull Requests

Categorical AI phenomenology: A first-person approach

Self-Speculation for Faster Reasoning Models

PrimeAgentOrchestrator: Memory-Primed Agent Spawning for Personal AI Infrastructure

Where Accountability Lives: Mapping Human Responsibility to Workflow Artifacts in Agentic Software Development

Twin: Playing an Unknown Game with a Test-Time Digital Twin

Token Reduction Is Not Cost Reduction

What Breaks When LLMs Code? Characterizing Operational Safety Failures of Agentic Code Assistants

Diagnosis Before Recovery: Turning Agent Failures into Selective Self-Correction

Principal Trait Analysis: Towards Deriving "Skills" in Human-AI Collaboration

Self-evolving network verifiers

FlexSQL: Flexible Exploration and Execution Make Better Text-to-SQL Agents

MEGA: Self-Evolving Agent Optimization Infrastructure via Wisdom Graph

One Recipe, Many Harnesses: What Self-Evolution Encodes Across Languages and Models