This project involves building a neural network from scratch to perform image classification on the Fashion MNIST dataset. The neural network is implemented in Python using NumPy and includes a double-layer architecture with sigmoid activation functions.
The project's main objective is to create a simple neural network without using external deep learning libraries. It demonstrates the fundamental concepts of neural networks, including forward and backward passes, weight initialization, activation functions, loss functions, and gradient descent.
The neural network consists of the following components:
- Weight Initialization: Xavier initialization for weight matrices.
- Activation Function: Sigmoid activation function.
- Loss Function: Mean Squared Error (MSE) for regression.
- Optimization: Stochastic Gradient Descent (SGD).