Repository for final project of Data Mining Course.
Individual project consisting in building a classifier for adversarial perturbation detection.
AY 2020/2021
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.
Complete description of assignment can be found here.
Presentation on proposed solution can be found here.
Source code of project can be found here.