Feedforward Neural Network from Scratch

Description

This project implements a Feedforward Neural Network (FFNN) from scratch, along with experimental notebooks that explore the impact of various configurations, including activation functions, weight initialization methods (Xavier and He), optimizers (SGD and Adam), regularization (L1 and L2), loss functions, and RMSNorm normalization. It also includes an extension using automatic differentiation (AdFFNNModel) and additional activation functions such as Leaky ReLU, ELU, SELU, and Softplus. The model’s performance is further evaluated against the MLPClassifier from scikit-learn as a baseline.

Contributors

NIM	Name
13523073	Alfian Hanif Fitria Yustanto
13523083	David Bakti Lodianto
13523091	Carlo Angkisan