This repo contains implementations of two mini-projects that were a part of Deep Learning course at EPFL.
The task of the first mini project is to implement different architectures of deep convolutional networks using PyTorch. By exploring different models we appreciated the effect of weight sharing and auxiliary losses on the final accuracy.
The task of the second mini project is to implement from scratch PyTorch tensor operations along with various DL layers with disabled autograd function. We tested the model by generating a "toy" dataset and showed that the created framework is working.
The repo also contains corresponding reports.