Neural networks: letter recognition

This project is a first part of research devoted to the implementation of neural networks into the bifurcation analysis of nuclear reactors. The focus is on the new concept - Neural Ordinary Differential Equation - and its superiority over the classical NN, such as multilayer perceptron. The comparison is built around ф well-studied problem of finding differences between two types of images. Datasets were formed through collecting individual handwriting from over 10 people, and then splitting them into train and test samples in the binarized form.

Multilayer Perceptron

Neural Ordinary Differential Equaltion