This repository contains the source code and the data of the term project "Deeper Hyperion" for the course CS454, KAIST
This work originated from the paper "DeepHyperion: Exploring the Feature Space of Deep Learning-Based Systems through Illumination Search" (Zohdinasab, Tahereh, et al. ISSTA 2021)
Click here to see how to set up DeepHyperion and validate its general functionality by using our pre-configured Docker image.
You may copy our version to the contatiner to run it without any setup.
Click here to see how to validate the paper’s claims and results in detail. This section introduces also extra use case scenarios to suggest how to extend the code beyond the scope of the paper.
The package is structured as follows:
- DeepHyperion-MNIST contains the diff-behavior DeepHyperion tool adapted to the MNIST dataset;
- DeepHyperion-FMNIST contains the DeepHyperion tool adapted to the Fashion-MNIST(FMNIST) dataset;
- experiments contains the raw experimental data and the scripts to obtain the results reported in the paper;
- documentation contains a quick installation guide and a detailed description of the tool.
Note: each sub-package contains further specific instructions.
The software we developed is distributed under MIT license. See the license file.
For any related question, please contact its authors:
- Subeom Lee ([email protected])
- Arogya Kharel ([email protected])
- Xiangchi Song ([email protected])
- Hyunjoon Cho ([email protected]).