Recognize Handwritten Letters

Description

The goal of this project is to recognize handwritten letters.

Dataset

We are working on a MNIST-like dataset for letters.

Above are shown the 25 first letters of the dataset.

Gallery

Learning curves & correlation matrix with a CNN network

Our CNN achieved more than 98% of accuracy on the test set.

Embedding of the letters with UMAP

Here we reduced the dimension of the dataset from 784 dimensions to 2, with the UMAP algorithm.