C++ Handwritten Digit Neural Network Classifier from Scratch with GUI

Engineered a complete neural network framework in native C++ supporting Linear, ReLU, and Softmax layers.

Intentionally designed to be modular and object-oriented for scalability, future work consists of adding new layer types and tools.

Implemented model serialization and an SFML interactive Graphical User Interface for digit drawing and prediction.

I designed this library to operate with a training config to specify training hyperparameters. The framework supports fully functional forward and backward propagation with a gradient descent optimizer.

Trained models can be loaded and saved seamlessly by writing model weights to a .txt file.