Neural Network Implementation in Java

This repository contains a basic neural network implementation written in Java. The project is currently under development and is intended for educational purposes, providing a clear and straightforward example of how to build a neural network from scratch without relying on external libraries.

Project Status

Status: Under Development 🚧

This project is still in the early stages of development. Features may be incomplete, and the implementation may change as it evolves.

Features

• Feedforward Neural Network: Basic implementation of a fully connected feedforward neural network.
• Backpropagation Algorithm: Initial implementation of the backpropagation algorithm for training the network.
• Activation Functions: Includes simple activation functions like tanh.

Since the project is under development, it is not yet packaged for installation. To use the code, you will need to clone the repository and manually compile and run the code.

Example

Here's a simple example of how to use the operations used in Defining a Neural Network:

        Value x1 = new Value(2.0);
        Value x2 = new Value(0.0);
        Value w1 =new Value(-3.0);
        Value w2 =new Value(1.0);
        Value b = new Value(6.88137358);
        Value x1w1 = x1.mul(w1);
        Value x2w2 = x2.mul(w2);
        Value x1w1x2w2 = x1w1.add(x2w2);
        Value n = x1w1x2w2.add(b);
        Value o = n.tanh();

Here's how you can define a small dataset and train it on a MLP of 1 input layer of 3 neurons, 2 hidden layer of 4 neuron each and 1 output layer with a single neuron with backpropagation:

        MLP n = new MLP(3, Arrays.asList(4, 4, 1));

        List<List<Double>> xs = Arrays.asList(
                Arrays.asList(2.0, 3.0, -1.0),
                Arrays.asList(3.0, -1.0, 0.5),
                Arrays.asList(0.5, 1.0, 1.0),
                Arrays.asList(1.0, 1.0, -1.0)
        );

        List<Double> ys = Arrays.asList(1.0, -1.0, -1.0, 1.0);

        for (int k = 0; k < 20; k++) {
            List<Value> ypred = new ArrayList<>();

            // Forward pass
            for (List<Double> x : xs) {
                List<Value> xValues = new ArrayList<>();
                for (Double xi : x) {
                    xValues.add(new Value(xi));
                }
                ypred.add(n.call(xValues).get(0));
            }

            Value loss = new Value(0.0);
            for (int i = 0; i < ys.size(); i++) {
                Value diff = ypred.get(i).sub(new Value(ys.get(i)));
                loss = loss.add(diff.pow(2));
            }

            for (Value p : n.parameters()) {
                p.setGrad(0.0);
            }

            // Backward pass
            loss.backward();

            // Update
            for (Value p : n.parameters()) {
                p.setData(p.getData() - 0.05 * p.getGrad());
            }

            System.out.println(k + " = " + loss.getData());
        }

Testing

There are currently no automated tests for this project. Manual testing is being done as features are developed.

Contributions

Contributions are welcome, but please note that the project is in an early development stage. If you'd like to contribute, feel free to fork the repository, create a new branch, and submit a pull request. Any suggestions or feedback are greatly appreciated!

Contact

For any questions, suggestions, or feedback, please contact:

• Email: manasbisht2507@gmail.com