In 1985, Rumelhart and others introduced back propagation and gradient descent as a method of training neural networks that have more than just an input layer and an output layer.
This Java project implements the algorithm described in Rumelhart's paper.
Be sure Java 11 and Maven are installed.
Copy the root directory and everything below it onto your machine.
Run "mvn clean test", and the library will be compiled and tests run.
The main class of the library is Network. A network is given a structure with "withStructure", and fed training patterns (that must match the input and output structure of the network) via "Network.learn()".
The NetworkTest.java file demonstrates proper (and improper) use.
Core Java.
"Learning Internal Representations by Error Propagation"
- DE Rumelhart, GE Hinton, RJ Williams - 1985 - apps.dtic.mil