MNIST comparer
This project aims at building a binary classifier for automated comparison of 2 handwritten digits (MNIST dataset) with the use of deep learning methods (Pytorch). It basically predicts which of the 2 digits is bigger.
Different architectures were compared, notably a siamese Neural Net structure using weight sharing, and 2 loss at different stage of the architectures (called auxiliary loss), please refer to the report PDF for more info.
Siamese architecture representation
This project was successfully completed in collaboration with Hila Vardi and Niccolò Stefanini, for the EPFL master class Deep learning by Prof. François Fleuret