Automated-Boundary-Testing-for-DL-Systems

Codebase for Giorgi Merabishvili's internship project on Neural Model-Based Test Generation for Deep Learning Systems.

Installation

conda env create -f environment.yml
conda activate mimicry

Getting started

1. Download pre-trained networks *.pkl files from Huggingface and put them under ./checkpoints/checkpoints.

2. Select one of the following datasets and go to the corresponding directory:

• MNIST: ./mnist
• FashionMNIST: ./f-mnist
• SVHN: ./svhn

3. Adjust the default config in ./<DATASET>/config.py.

4. Run the file ./<DATASET>/search.py to generate frontier pairs.

Acknowledgement

This code is developed based on StyleGAN3 and PTI