adversarial-nn

This project was executed as a school assignment at the University of Twente. In this project a basic CNN model (created by ourselves) and the ResNet-50 model are trained upon the Fashion-MNIST dataset whereafter different adversial attacks and defences are applied to look at their impact on the classification accuracy.

Project Overview

• School: University of Twente
• Course: Deep Learning - From Theory to Practice
• Assignment Type: Topic wiht open implementation
• Group Size: 4

Execution

All code with explanation can be found in the main.ipynb notebook. Trained models are saved in the model directory and figures are saved in the results dictionary.