.png){kind=icon}
.png){kind=icon}
Return
SemuFuzz: An efficient fuzzing approach for deep learning libraries based on semantic representation
.png){kind=icon}
PRE
AI
Z
陈
S
J
W
DOI：10.1016/j.jss.2026.112999.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Proposes SemuFuzz, a semantic-aware fuzzing framework that models API constraints from documentation to guide edge-case generation for deep learning libraries. • Combines lightweight semantic embeddings with logical predicates to enable reusable, cross-API test inputs that significantly improve coverage and seed validity. • Discovers 36 crashes, including 10 previously unknown vulnerabilities in PyTorch and TensorFlow; two have been confirmed and patched by developers.
Journal
IF:
4.1
Papers:
5.4K
Citations:
8.4K
Organization
Cited Papers