A deep learning framework

This project implements a deep learning framework using only pytorch Tensors. In particular, we implemented forward and backward passes for the MSE loss and for the most common operations used within a neural network, e.g. fully connected layer, ReLU activation. The the framework is versatile and easy to understand which allow to easily develop additional features.

Folders and files:

• hotgrad/: folder containing our deep learning framework.
• dataset_generator.py: function used to generate a simple 2D dataset.
• test.py: tests the neural network implemented with our library and shows the accurancy.
• Report.pdf: report explaining our approach and implemenation of the deep learning library.

Report

For a detailed description of the implemented solution please refer to the Report.