Neural Network from Scratch

This project implements a neural network from scratch in pure Java 21. The neural network is capable of training and testing using the MNIST dataset for digit recognition.

Features

• Neural Network Implementation: The core of the project is the neural network implementation found in NeuralNetwork.
• Training and Testing: The network can be trained and tested using the MNIST dataset. The training is done using stochastic gradient descent, implemented in StochasticGradientDescentTrainer.
• Cost Functions: Different cost functions are available, such as the cross-entropy cost function in CrossEntropyCostFunction and the quadratic cost function in QuadraticCostFunction.
• Activation Functions: The project includes various activation functions, such as the sigmoid function in SigmoidFunction.
• Regularization: Regularization techniques like L1 and L2 regularization are implemented in L1Regularization and L2Regularization.
• Data Handling: Utilities for handling data, such as shuffling and subdividing datasets, are provided in DataUtils.
• MNIST Data Loader: The MNIST dataset loader is implemented in MNISTLoader.
• GUI: A graphical user interface for drawing digits and interacting with the neural network is implemented in Frame, ButtonPanel, and DrawPanel.

Getting Started

Prerequisites

• Java 21
• Maven

Building the Project

To build the project, run the following command in the root directory:

mvn clean install

mvn exec:java -Dexec.mainClass="org.neural.network.testermodule.Frame"