MNISTNeuralNetwork

A multi-layer (3 layer) fully connected, feed forward network consisting of sigmoidal neurons, trained with stochastic gradient descent and back propagation.

Documentation

Overview

The MNISTNeuralNetwork program is a simple artificial neural network implemented in Java to classify handwritten digits from the MNIST dataset. It includes functionalities to train the network, load pre-trained weights, test its accuracy, and visualize misclassified images.

Features

• Train the neural network using mini-batches and backpropagation.
• Load and save trained weights to a file.
• Display network accuracy on both training and testing datasets.
• Visualize misclassified testing images.
• User-interactive console menu for easy navigation.

Neural Network Architecture

• Input Layer: 784 neurons (corresponding to 28x28 pixel grayscale images).
• Hidden Layer: 100 neurons.
• Output Layer: 10 neurons (one for each digit from 0 to 9).
• Activation Function: ReLU for hidden layers, Softmax for the output layer.
• Learning Rate: 0.01.
• Mini-batch Size: 32.
• Epochs: 50.

Program Execution

Upon running the program, the user is prompted to choose between:

1. Training the network from scratch.
2. Loading a pre-trained network.
3. Exiting the program.

After choosing an option, additional functionalities become available, including:

• Retraining the network.
• Displaying accuracy.
• Running the network on test data.
• Viewing misclassified images.
• Saving the network state.

Core Functions

feedforward(double[] input)

Computes the output of the network given an input image.

backpropagation(double[] input, double[] target)

Performs the backpropagation algorithm to adjust weights and biases based on error.

updateweightsandbiases(double[] input)

Updates network weights and biases after computing errors.

train(String trainingDataPath, int trainingSize)

Trains the network using the specified dataset.

accuracy(String dataPath, int dataSize)

Calculates the accuracy of the network on a given dataset.

saveWeightsToFile(String filename)

Saves the trained weights and biases to a file for future use.

loadWeightsFromFile(String filename)

Loads saved weights and biases from a file.

runOnTestDataAndDisplay(String dataPath, int dataSize)

Runs the network on testing data and displays the results, including misclassified images.

File Handling

• Training and testing data should be in CSV format.
• Weight sets are stored in weightset.csv.

Dependencies

• Java Standard Library (java.io, java.util)

Usage Example

javac MNISTNeuralNetwork.java
java MNISTNeuralNetwork

Follow on-screen prompts to train, load, and test the network.