LeNet MNIST

This project is a simple implementation of the LeNet architecture, written in PyTorch, proposed in: LeCun, Y.; Bottou, L.; Bengio, Y. & Haffner, P. (1998). Gradient-based learning applied to document recognition. The network is trained on the MNIST handrawn numbers dataset.

Loss function

The loss function of choice is the Cross Entropy loss function

Optimizer

The algorithm used for optimization is the Adam optimizer

Architecture

Image comes from wikipedia: https://en.wikipedia.org/wiki/LeNet#/media/File:Comparison_image_neural_networks.svg

The code implements the architecture highlighted in LeNet part of the image.

Results

This architecture reached on an M2 Pro 97.3% accuracy on the MNIST dataset