This project was created as part of the Neural Networks course at Warsaw University of Technology.
The goal of the project was to implement the Multi-Layer Perceptron Neural Network from scrach without use of any advanced Deep Learning libraries, implementing own forwards and backwards propagation methods.
The Neural Network is capable of handling batched, vectorized calculations and solving classification as well as non-linear regression problems as shown in example files.
Model trained on MNIST after 60 epochs scores 97% accuracy without use of any augmentation or dropout methods.
Dawid Kożykowski