We propose CertPri, a test input prioritization technique designed based on a movement cost perspective of test inputs in DNNs’ feature space. CertPri differs from previous works in three key aspects: (1) certifiable - it provides formal robustness guarantee for the movement cost; (2) effective - it leverages formal guaranteed movement costs to identify malicious bug-revealing test inputs; and (3) generic - it can be applied to various tasks, data forms, models, and scenarios. CertPri significantly improves 53.97% prioritization effectiveness on average compared with baselines. Besides, its robustness and generalizability are 1.41-2.00 times and 1.33-3.39 times that of baselines on average, respectively.
A technical description of CertPri is available in this paper. Below is the bibtex entry for this paper.
@inproceedings{Zheng2023CertPri,
author = {Zheng, Haibin and Chen, Jinyin and Jin, Haibo},
title = {CertPri: Certifiable Prioritization for Deep Neural Networks via Movement Cost in Feature Space},
booktitle = {38th IEEE/ACM International Conference on Automated Software Engineering},
address = {Belval, Esch-sur-Alzette, Luxembourg},
pages = {1-13},
date = {September 11 - 15},
publisher = {{IEEE/ACM}},
year = {2023}
}