Deep Learning

• Project 1 : Image classification,weight sharing, auxilliary losses

The objective of this project is to test different architectures to compare two digits visible in a two-channel image. It aims at showing in particular the impact of weight sharing, and of the use of an auxiliary loss to help the training of the main objective.

• Project 2 : Mini Deep-Learning framework

The objective of this project is to design a mini “deep learning framework” using only pytorch’s tensor operations and the standard math library, hence in particular without using autograd or the neural-network modules.

Data set

• MNIST
• Set of 1,000 points

Authors

• Sajaendra Thevamanoharan
• Deniz Sungurtekin
• Antoine Branca