MiaoDX/bp_java

A sample back propagation implementation using java.

% A simple implementation of Backpropagation in Java. % MiaoDX MiaoDX@hotmail.com % Oct, 2016

What we want to do.

There are lots of amazing open source implements of ML(/DL) libraries,and before diving into these great porjects,I think it really necessary to implement a simple ANN just by hand.In this way,we will not treat these toolkits as BlackBoxes and just Draw a graph and hope that we will get nice performace after an uncertain time,on the other hand,we make sure that we CAN write an ANN or a simple part of it.

In this repo,a java implementation of Backpropagation is presented.

Reference

There are some great resources on the internet that explains BP,some clear,some mathematical,and some just too difficult or too complex for us(me especially) to understand.Here are some articals I think worth reading:

1. 神经网络反向传播的数学原理 [CHINESE]

2. A Gentle Introduction to Artificial Neural Networks

3. Principles of training multi-layer neural network using backpropagation

4. BP（Back Propagation）神经网络及Matlab矩阵实现 [CHINESE]

At first,I wrote my codes according to the third article,but later I found that chances are that when calc the error($\delta$),it can be something wrong,and then I found the second one,It's awesome.But I still assist you read the third artical(first),for it's really expressive.

Design and implementation

To not make this page too complex,THE DESIGN PRINCIPLES AND IMPLEMENTATIONS DETAILS OR TRICKS CAN BE FOUND IN doc/DesignAndImplement.md,IT IS A GOOD ARTICLE -.-.

ChangeLog

TODO: USE Guava.Graph to reimplement it.

20151026: Some visualization work.

20161025: An implement do things ok.

20161004: Basic skeleton.

Good Luck & Have Fun. -- miaodx