The project focuses on the training of a Deep Belief Network using various hyperparameters such as hidden layers, nodes, and optimizer choices for optimizing performance in image classification. It analyzes the internal representations developed by the model through hierarchical clustering and feature visualization techniques. The goal is to understand the accuracy and psychometric curves for the DBN. Finally, it compares the ability of two models—the feedforward network and the DBN—to resist adversarial attacks.
Image Classification, Deep Belief Network (DBN), FashionMNIST, Adversarial Attacks
- FashionMNIST dataset
- Deep Belief Network
- Feedforward networks
- Python (Pytorch)
- Code:
code.ipynb
: Jupyter notebook that run al the project.DBN.py
andRBN.py
: Have the functions called bycode.ipynb
.