guidodc97 / adversarialperturbationdetection

Repository for final project of Data Mining Course

Adversarial Perturbation Detection

Repository for final project of Data Mining Course.

Individual project consisting in building a classifier for adversarial perturbation detection.

AY 2020/2021

Solution

Project has been completely developed in Python, using the popular framework for deep learning Pytorch.

The steps followed to solve the problem are:

• Data preparation and analysis
• Modeling
  • Model selection
  • Hyperparameter tuning
• Model evaluation

The solution chosen is AlexNet architecture trained using transfer learning technique, which gave an accuracy of 98.7% on test set.

Documentation & Code

Complete description of assignment can be found here.

Presentation on proposed solution can be found here.

Source code of project can be found here.