{
  "added_at": "2026-10-20",
  "arxiv_id": "2606.16162",
  "arxiv_version_reviewed": "v2",
  "authors": [
    "Peipei Liu",
    "Jian Sun",
    "Mingzhe Xing",
    "Yicheng Zeng",
    "Zhaoteng Yan",
    "Lixiao Zhang",
    "Li Chen",
    "Dan Li"
  ],
  "bibtex_url": "https://awesomerea.org/papers/citations/paper-autodecompiler.bib",
  "code_url": "https://zenodo.org/records/23120286",
  "date_basis": "First arXiv submission on 2026-06-15; the reviewed v2 revision is 2026-10-03, not a separate publication.",
  "date_basis_zh": "采用 2026-06-15 的 arXiv 首投日期；本次核对的 v2 修订日期为 2026-10-03，并非另一次出版。",
  "date_precision": "day",
  "description_en": "AutoDecompiler trains function-level decompilation models to revise assembly- or pseudocode-derived outputs using compilation, execution and I/O feedback. Its reinforcement-learning design combines diagnostic feedback with rewards for valid code, tested behavior, syntax and data-flow structure.",
  "description_zh": "AutoDecompiler 训练函数级反编译模型，利用编译、执行及 I/O 反馈修正汇编或伪代码输入产生的输出。其强化学习设计结合诊断反馈及代码有效性、测试行为、语法和数据流结构奖励。",
  "doi": null,
  "entry_url": "https://awesomerea.org/papers/paper-autodecompiler/",
  "entry_url_zh": "https://awesomerea.org/zh/papers/paper-autodecompiler/",
  "evidence_urls": [
    "https://arxiv.org/abs/2606.16162v2",
    "https://arxiv.org/html/2606.16162v2",
    "https://arxiv.org/abs/2606.16162v1",
    "https://arxiv.org/pdf/2606.16162v2",
    "https://zenodo.org/api/records/23120286",
    "https://huggingface.co/api/models/AutoDecompiler/AutoDecompiler-1.3B-e2e"
  ],
  "first_arxiv_date": "2026-06-15",
  "id": "paper-autodecompiler",
  "json_url": "https://awesomerea.org/papers/paper-autodecompiler/record.json",
  "kind": "research",
  "markdown_url": "https://awesomerea.org/papers/paper-autodecompiler/index.md",
  "markdown_url_zh": "https://awesomerea.org/zh/papers/paper-autodecompiler/index.md",
  "metadata_notes": "First submitted 2026-06-15; reviewed v2 is dated 2026-10-03. The arXiv record uses the shorter catalog title, while the v2 PDF/HTML use a longer title beginning “AutoDecompiler: Reinforcement-Learning-Trained”; eight authors match. Zenodo version 23120286 lists code/data archives under CC BY 4.0. The inspected Hugging Face model has no license metadata.",
  "metadata_notes_zh": "首投日期为 2026-06-15，本次核对的 v2 日期为 2026-10-03。目录采用 arXiv 记录的较短标题，v2 PDF／HTML 采用以“AutoDecompiler: Reinforcement-Learning-Trained”开头的更长标题，八位作者一致。Zenodo 版本 23120286 列出采用 CC BY 4.0 的代码／数据归档。已查 Hugging Face 模型未提供许可元数据。",
  "pdf_url": "https://arxiv.org/pdf/2606.16162v2",
  "project_url": "https://huggingface.co/AutoDecompiler",
  "publication_date": "2026-06-15",
  "publication_status": "preprint",
  "record_type": "paper",
  "runtime_tested": false,
  "scope_limitations_en": "Evaluation uses C adaptations of HumanEval and MBPP with validation feedback; finite tests do not establish general equivalence. HumanEval also informs reward-weight selection, and base-model pretraining exposure cannot be ruled out. Refinement is context-limited, whole-program dependencies are not recovered, and unseen toolchains or protected binaries remain limitations.",
  "scope_limitations_zh": "评估采用 HumanEval、MBPP 的 C 改编版本及验证反馈，有限测试不证明一般等价。HumanEval 还用于选择奖励权重，无法排除基座模型预训练时接触基准。修正受上下文长度限制，不恢复完整程序依赖，对未见工具链或受保护二进制仍有局限。",
  "targets": [
    "binary"
  ],
  "title": "Binary Decompilation LLM with Feedback-Driven Multi-Turn Refinement",
  "title_zh": "基于反馈驱动多轮修正的二进制反编译大语言模型",
  "url": "https://arxiv.org/abs/2606.16162",
  "venue": "arXiv",
  "verification": "source-reviewed",
  "verified_at": "2026-10-20",
  "why_fits_en": "Adds reinforcement-learning-trained use of program feedback, with separately recorded first submission and latest revision.",
  "why_fits_zh": "补充通过强化学习学习利用程序反馈的方案，并分别记录首投与最新修订。",
  "year": 2026
}