paperbot · PL 论文追踪

全部论文 待读 收藏

P2 A comprehensive study on deep learning bug characteristics

深度学习缺陷特征的全面研究
FSE Proceedings of the 2019 27th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering2019匹配度 3/10 · 该论文是深度学习库缺陷的实证研究，与用户关注的 ML 系统静态分析有间接关联：其缺陷类型、根因和反模式分析可为分析/验证框架提供动机。但它不涉及静态分析、Agentic 程序、LLM 辅助分析、浮点程序分析或贝叶斯程序分析，因此与核心研究方向匹配度较低。引用 9 inbox
Md Johirul Islam, Giang Nguyen, Rangeet Pan, Hridesh Rajan
机器学习与程序语言测试、调试与修复
crossref: 2026-09-12

AI 速览

一句话：通过 Stack Overflow 与 GitHub 数据实证刻画深度学习库缺陷的类型、根因、影响与反模式。

深度学习开发者常依赖库和工具构建应用，但对其软件缺陷特征缺乏系统理解。本文分析 Stack Overflow 的 2716 条高质量帖子，以及 GitHub 上 Caffe、Keras、TensorFlow、Theano 和 Torch 五个流行库的 500 个缺陷修复提交，考察缺陷类型、根因、影响、流水线易发阶段和常见反模式。结果显示，数据缺陷与逻辑缺陷是最严重的缺陷类型，出现频率超过 48%；主要根因是错误模型参数（IPS）和结构低效（SI），出现频率超过 43%。研究还发现库使用中存在一些常见反模式，可指导更好的平台、调试机制、开发实践及分析验证框架。

问题
深度学习软件依赖库和工具，但社区对其缺陷类型、根因、影响、易发阶段与反模式缺乏系统认识。
方法
对 Stack Overflow 2716 条高质量帖子和 GitHub 上五个深度学习库（Caffe、Keras、TensorFlow、Theano、Torch）的 500 个缺陷修复提交进行实证分析。
评估
通过统计与分类这些帖子/提交，报告缺陷类型、根因与影响的频率，并总结常见反模式。
原文摘要（Abstract）

Deep learning has gained substantial popularity in recent years. Developers mainly rely on libraries and tools to add deep learning capabilities to their software. What kinds of bugs are frequently found in such software? What are the root causes of such bugs? What impacts do such bugs have? Which stages of deep learning pipeline are more bug prone? Are there any antipatterns? Understanding such characteristics of bugs in deep learning software has the potential to foster the development of better deep learning platforms, debugging mechanisms, development practices, and encourage the development of analysis and verification frameworks. Therefore, we study 2716 high-quality posts from Stack Overflow and 500 bug fix commits from Github about five popular deep learning libraries Caffe, Keras, Tensorflow, Theano, and Torch to understand the types of bugs, root causes of bugs, impacts of bugs, bug-prone stage of deep learning pipeline as well as whether there are some common antipatterns found in this buggy software. The key findings of our study include: data bug and logic bug are the most severe bug types in deep learning software appearing more than 48% of the times, major root causes of these bugs are Incorrect Model Parameter (IPS) and Structural Inefficiency (SI) showing up more than 43% of the times.We have also found that the bugs in the usage of deep learning libraries have some common antipatterns.

链接与引用

DOI 原文 · PDF（开放获取） ·

BibTeX 
@inproceedings{paperbot5078,
  title = {A comprehensive study on deep learning bug characteristics},
  author = {Md Johirul Islam and Giang Nguyen and Rangeet Pan and Hridesh Rajan},
  booktitle = {FSE},
  year = {2019},
  doi = {10.1145/3338906.3338955}
}