Effective Python-frontend fuzzing for deep learning libraries with runtime coverage feedback

doi:10.1016/j.infsof.2026.108195

Effective Python-frontend fuzzing for deep learning libraries with runtime coverage feedback

2026-05-18

0

PRE

AI

DOI：10.1016/j.infsof.2026.108195

Original

Original request for help

Save

Abstract

En 中文

In recent years, fuzzing techniques for deep learning libraries have been proposed. Although these techniques can detect vulnerabilities, they exhibit notable limitations. Existing methods either rely on blackbox testing mechanisms to generate inputs, leading to limited input effectiveness, or only provide runtime coverage feedback through the C++ frontend, while overlooking vulnerability detection in the Python frontend, which represents the primary usage scenario.

Keywords:

fuzzing

deep learning libraries

Python frontend

runtime coverage feedback

vulnerability detection

Effective Python-frontend fuzzing for deep learning libraries with runtime coverage feedback

Abstract

Journal

Organization