Neural Network Digit Classifier

A neural network built from scratch in Python to classify handwritten digits for CS 70, Foundations of Applied Computer Science and Machine Learning. The implementation includes forward computation and backward propagation for linear and non-linear ReLU layers, as well as dropout regularization. The model was trained using gradient descent with least squares loss.

A Bayesian hyperparameter optimizer was designed to tune the model's architecture and training parameters. Achieved the highest test accuracy of 92.5% in the class, winning a citation of merit among 150 CS majors at Dartmouth.