{
  "added_at": "2026-10-20",
  "arxiv_id": "2605.15334",
  "arxiv_version_reviewed": "v2",
  "authors": [
    "Yihong Dong",
    "Jiaru Qian",
    "Haoran Zhang",
    "Peixu Wang",
    "Binhua Li",
    "Zhi Jin",
    "Yongbin Li",
    "Ge Li",
    "Xiaokang Yang",
    "Xue Jiang"
  ],
  "bibtex_url": "https://awesomerea.org/papers/citations/paper-dio-agent.bib",
  "code_url": "https://github.com/JiaruQian/IO2Code",
  "date_basis": "Official NeurIPS 2026 poster listing confirms acceptance, not a proceedings publication date. First arXiv submission was 2026-05-14; reviewed v2 was revised on 2026-10-18.",
  "date_basis_zh": "NeurIPS 2026 官方海报页面确认接收，不据此认定论文集出版日期。arXiv 首投为 2026-05-14；所审阅 v2 修订于 2026-10-18。",
  "date_precision": "year",
  "description_en": "Synthesizes executable programs from input/output observations using LLM-guided evolutionary search, progressively revealed examples, simple-to-complex mutations and execution-error feedback. An active variant chooses inputs and queries a hidden reference program.",
  "description_zh": "通过 LLM 引导的演化搜索、逐步展示的样例、由简到繁的修改及执行错误反馈，从输入输出观察合成可执行程序；主动变体自行选择输入并查询隐藏的参考程序。",
  "doi": "10.48550/arxiv.2605.15334",
  "entry_url": "https://awesomerea.org/papers/paper-dio-agent/",
  "entry_url_zh": "https://awesomerea.org/zh/papers/paper-dio-agent/",
  "evidence_urls": [
    "https://arxiv.org/abs/2605.15334",
    "https://arxiv.org/html/2605.15334v2",
    "https://api.datacite.org/dois/10.48550/arxiv.2605.15334",
    "https://neurips.cc/virtual/2026/loc/sydney/poster/154711",
    "https://github.com/JiaruQian/IO2Code/blob/ef15827ab7090bf2d227fb1eef51d91b0163339e/README.md",
    "https://github.com/JiaruQian/IO2Code/tree/ef15827ab7090bf2d227fb1eef51d91b0163339e"
  ],
  "first_arxiv_date": "2026-05-14",
  "id": "paper-dio-agent",
  "json_url": "https://awesomerea.org/papers/paper-dio-agent/record.json",
  "kind": "research",
  "markdown_url": "https://awesomerea.org/papers/paper-dio-agent/index.md",
  "markdown_url_zh": "https://awesomerea.org/zh/papers/paper-dio-agent/index.md",
  "metadata_notes": "The official poster and pinned repository retain the earlier title From I/O to Code with Discovery Agent; all ten authors match v2 and DataCite. The registered DOI identifies the arXiv work. The paper uses the arXiv non-exclusive license. Author-linked code was checked at ef15827ab7090bf2d227fb1eef51d91b0163339e: its README documents Python 3.10+, provider credentials, benchmark/oracle tasks and experiment runners. No repository-wide code/data license was verified in the reviewed README and root tree.",
  "metadata_notes_zh": "官方海报与固定仓库保留早期标题 From I/O to Code with Discovery Agent，十位作者与 v2 及 DataCite 一致。已注册 DOI 标识 arXiv 工作。论文采用 arXiv 非独占许可。作者链接代码固定核查于 ef15827ab7090bf2d227fb1eef51d91b0163339e；README 说明 Python 3.10+、服务商凭据、基准／参考程序任务和实验入口。所审阅 README 与根目录树未核实覆盖整个仓库的代码／数据许可。",
  "pdf_url": "https://arxiv.org/pdf/2605.15334v2",
  "project_url": null,
  "publication_date": null,
  "publication_status": "conference-accepted",
  "record_type": "paper",
  "runtime_tested": false,
  "scope_limitations_en": "IO2CodeBench has 125 generated tasks spanning basic operations, algorithms, geometry and image-to-value mappings, normally with eight visible and fifteen held-out examples. Active exploration covers a 25% subset from each level. Multimodal results average sample pass ratios, while other levels count tasks passing every held-out example. Finite test success does not establish original-source recovery or complete equivalence; arbitrary applications and stateful services are not evaluated.",
  "scope_limitations_zh": "IO2CodeBench 包含 125 个生成任务，覆盖基础运算、算法、几何及图像到数值的映射，通常每题提供八个可见样例和十五个留出样例。主动探索只覆盖各层级的 25% 子集。多模态结果采用样例通过比例的均值，其他层级按通过全部留出样例的任务计数。有限测试通过不代表恢复原源码或完全等价；未评测任意应用及有状态服务。",
  "targets": [
    "app"
  ],
  "title": "Agentic Reverse Engineering: From I/O to Code with Discovery Agent",
  "title_zh": "智能体逆向工程：利用发现智能体从输入输出行为恢复代码",
  "url": "https://arxiv.org/abs/2605.15334",
  "venue": "NeurIPS 2026",
  "verification": "source-reviewed",
  "verified_at": "2026-10-20",
  "why_fits_en": "Provides a concrete observation-to-program reconstruction workflow and separates visible-example fitting from held-out behavior checks.",
  "why_fits_zh": "提供从行为观察到程序重建的具体流程，并区分可见样例拟合与留出行为检查。",
  "year": 2026
}