Multilayer Networks (MLN) are easy to understand but need deep machine learning knowledge to master. The edu.yaprnn app helps you learn MLNs by letting you experiment with sample data, network settings, and training parameters.

In the app, you create and train MLNs using MNIST digit images and vocal sound files.

Our team of 5-6 students developed the original version for a machine learning course years ago. I updated it to fix old issues. The original version is at the yaprnn repository, and a minimally fixed version is on the old-yaprnn branch in this repository.

• Have a Java 21+ distribution installed, e.g. Eclipse Adoptium.
• Exec gradlew run on a terminal to run the java application.

The app is divided into two sets of packages. Most of it is for the GUI, found in gui, events, model, and support. If you're interested in the MLN code, check out the networks package and the related functions, samples, and training packages.

C4Component
    title edu.yaprnn

    Container_Boundary(yaprnn_mln, "edu.yaprnn Multilayer Networks") {
        Component(networks, "Networks", "edu.yaprnn.networks", "Creates and trains multilayer networks.")
        Component(functions, "Activation Functions", "edu.yaprnn.functions", "Provides activation functions.")
        Component(samples, "Samples", "edu.yaprnn.samples", "Provides samples and labels for classification.")
        Component(training, "Training", "edu.yaprnn.training", "Organizes samples into training and test lists.")
    }

    Container_Boundary(yaprnn_gui, "edu.yaprnn GUI") {
        Component(gui, "GUI", "edu.yaprnn.gui", "The editor and training tool for multilayer networks.")
        Component(events, "Events", "edu.yaprnn.events", "Handles app events for the GUI.")
        Component(model, "Model", "edu.yaprnn.model", "Manages samples, training lists, network templates, and networks.")
        Component(support, "Support", "edu.yaprnn.support", "Includes configuration and utility tools.")
    }

edu.yaprnn is published under Creative Commons 4.0 (CC BY-NC). It uses open source software published under multiple different licenses.

The repository contains the MNIST database, which consists of 60.000 digit samples.

Attributions for flaticon graphics:

• Buandesign
• Mayor Icons
• Vectorslab
• srip
• afif fudin