FAST: Boosting Uncertainty-based Test Prioritization for Neural Networks via Feature Selection (ASE 2024)

See the ASE 2024 paper for more details.

Prerequisite (Py3.6 & Tf2)

The code are run successfully using Python 3.6 and Tensorflow 2.6.0.

We recommend using conda to install the tensorflow-gpu environment

conda create -n tf2-gpu tensorflow-gpu==2.6.0
conda activate tf2-gpu

Checking installed environments

conda env list

Files

• metrics: metrics for evaluating the test prioritization task.
• coverage: neuron-coverage-based prioritization methods.
• suprise: surprise-adequacy-based prioritization methods.
• uncertainty: uncertainty-based prioritization methods.
• posthoc: post-hoc methods for uncertainty-based cerprioritization.

Acknowledgements

FAST has been partially developed with the support of European Union’s ELSA – European Lighthouse on Secure and Safe AI, Horizon Europe, grant agreement No. 101070617.

Publication

@inproceedings{fast,
author = {Jialuo Chen, Jingyi Wang, Xiyue Zhang, Youcheng Sun, Marta Kwiatkowska, Jiming Chen, Peng Cheng},
title = {FAST: Boosting Uncertainty-based Test Prioritization for Neural Networks via Feature Selection},
booktitle = {39th IEEE/ACM International Conference on Automated Software Engineering, ASE 2024, California, United States, October 27- November 1, 2024},
year = {2024},
}