paperbot · PL 论文追踪

全部论文 待读 收藏

P1 Knowledge-Based Version Incompatibility Detection for Deep Learning

面向深度学习的基于知识的版本不兼容检测
FSE Proceedings of the 31st ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering2023匹配度 5/10 · 论文属于机器学习系统与程序语言/软件工程交叉，关注深度学习项目版本不兼容检测，并用预训练QA模型抽取Stack Overflow知识辅助检测，与“PL+AI”有弱关联；但它不是静态分析、Agentic程序、浮点或贝叶斯程序分析，也未使用当前强LLM/Agent，因此不是核心方向。引用 6 inbox
Zhongkai Zhao, Bonan Kou, Mohamed Yilmaz Ibrahim, Muhao Chen, Tianyi Zhang
机器学习与程序语言编程环境与开发者工具
crossref: 2026-09-12dblp: 2026-09-12arxiv: 2026-09-12

AI 速览

一句话：提出用预训练问答模型从 Stack Overflow 抽取版本兼容知识并构建加权知识图，以检测深度学习项目中的版本不兼容。

论文针对深度学习模型与应用复用/复现中频繁出现的版本不兼容问题。现有方法主要依赖 PyPI 中声明的库依赖，难以发现未文档化版本约束，也无法处理硬件驱动或操作系统相关版本问题。作者利用 Stack Overflow 的版本问题讨论，将知识抽取建模为问答任务，用预训练 QA 模型抽取版本兼容知识，并整合为加权知识图，以在复用深度学习项目时检测潜在不兼容。评估显示，知识抽取准确率为 84%；在 10 个流行深度学习项目中能识别 65% 已知版本问题，精确率 92%，而两个现有最佳方法仅检测 29% 和 6%，精确率分别为 33% 和 17%。

问题
复用或复现深度学习模型/应用时，版本不兼容问题普遍；现有技术仅能利用 PyPI 声明的库依赖，无法覆盖未文档化约束以及硬件驱动或 OS 相关版本问题。
方法
将版本兼容知识抽取形式化为问答任务，使用预训练 QA 模型从 Stack Overflow 讨论中抽取知识，并整合为加权知识图来检测潜在版本不兼容。
评估
在知识抽取上达到 84% 准确率；在 10 个流行深度学习项目中识别 65% 的已知版本问题且精确率 92%，显著优于两个 SOTA 方法（29%/6%，精确率 33%/17%）。
原文摘要（Abstract）

Version incompatibility issues are rampant when reusing or reproducing deep learning models and applications. Existing techniques are limited to library dependency specifications declared in PyPI. Therefore, these techniques cannot detect version issues due to undocumented version constraints or issues involving hardware drivers or OS. To address this challenge, we propose to leverage the abundant discussions of DL version issues from Stack Overflow to facilitate version incompatibility detection. We reformulate the problem of knowledge extraction as a Question-Answering (QA) problem and use a pre-trained QA model to extract version compatibility knowledge from online discussions. The extracted knowledge is further consolidated into a weighted knowledge graph to detect potential version incompatibilities when reusing a DL project. Our evaluation results show that (1) our approach can accurately extract version knowledge with 84% accuracy, and (2) our approach can accurately identify 65% of known version issues in 10 popular DL projects with a high precision (92%), while two state-of-the-art approaches can only detect 29% and 6% of these issues with 33% and 17% precision respectively.

链接与引用

DOI 原文 · arXiv · PDF（开放获取） · DBLP

BibTeX 
@inproceedings{abs-2308-13276,
  title = {Knowledge-Based Version Incompatibility Detection for Deep Learning},
  author = {Zhongkai Zhao and Bonan Kou and Mohamed Yilmaz Ibrahim and Muhao Chen and Tianyi Zhang},
  booktitle = {FSE},
  year = {2023},
  doi = {10.1145/3611643.3616364}
}