Skip to content
Java Artificial Neural Network Library
Java
Find file
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.
docs
gradle clean up
src
.gitignore gitignore
.project
COPYING
LICENSE
build.gradle
gradlew
gradlew.bat adding gradle build
readme.md Update readme.md
settings.gradle

readme.md

Development has been stopped as JANN has been superseded by Spark MLib

Introduction

JANN is an Artificial Neural Networks written in JAVA, licensed under Affero GPL V3. The project is in preAlpha phase "Unstable versions 0.x" and lots of changes are inevitable.

Technologies

  • Java
  • JBlas (Java + Fortran)
  • Gradle (groovy)

Updates

Currently moving to Spark engine, Spark uses mllib which uses JBlas; in the early versions of JANN, JBlas was used directly, now it will be used through spark. Stay tuned!

Non-Functional Requirements

High Performance

Jann performance is an essential non-functional requirement. The idea was originally CANN (C++ Artificial Neural Network). What differs Jann from many other Java ANN libraries is JBLAS a very promising Math library that dares to achieve very high speed of calculations and compete with C++ performance jblas is essentially a light-wight wrapper around the BLAS and LAPACK routines. These packages have originated in the Fortran community which explains their often archaic API. On the other hand modern implementations are hard to beat performance wise. jblas aims to make this functionality available to Java programmers such that they do not have to worry about writing JNI interfaces and calling conventions of Fortran code. Also in Jann I'm trying as much as possible to optimize the code while making it also readable.

Dynamic

Designing a network is no longer a problem. Neural networks comes in many different structures (competitive, feed forward and feed backward etc). So by using Graph Theory to structure the linkage between layers inside the network it self. Where the indices of layers are represented as the JGraph data structure vertex, you can connect any layers together; this way a training such as Back-propagation would be in its most easy and clear possible form and in easy steps you can know a layer's predecessor or successors.

Testimonials

SoftPedia.com JANN is a unique, open source, small, easy to use Artificial Neural Networks library.

History & Release Notes

Jann trials took tens of hours, I'm not a neural network expert or mathematician so I didn't have the required background, The moment I learn about an algorithm I run back to implement it in jann, but the more complex techniques get the more I become aware of similarities and ways to have more robust, general, and dynamic neural network training.

Before jann there was CANN a C++ version of Jann. Implementation of C++ applications is very powerful but it lacks the ease of deployment and update as you know that only for you to use a dynamic library (.SO) in linux would be a huge burden or if you tried to go through the MAKE FILE hassle. though I found operator overloading very powerful technique I also found that sometimes you can just be ok without it.

JANN-0.4:

Part 1 Linear regression One/Multiple Variables Linear regression Training: Batch GD (single neuron) Stochastic GD (single or nNeurons) Feature Mapping with degrees Feature Scaling with (STD, average) Normal Equation (single neuron) Dynamic Training through com.marmoush.jann.train.Train Class Better Tests

Part 2 Models:

  • Logistic Regression One Class
  • Logistic Regression Multiple Classes

Training:

  • Logistic Regression GD
  • Regularized Linear Regression
  • Regularized Logistic Regression

Utils:

  • Random Initialization of theta

Part 3

Models: 1- Logistic Regression Neural Network 2- Linear Regression Neural Network

Gradient Checking Numerical Estimation of gradients Neural Network Back propagation

Part 4 Notes or Logic for Decision Making For handling errors :

  1. Identify Problems with the help of (Graphs, tools)
  2. Act upon on it (Manual ways, Automatic ways)

Part 5

  • More Tests
  • More Graphs

JANN-0.3

  • Graphed Network
  • Functors IFunctionable aka Java function pointers {IPerformance, Iweight, ITransfere}
  • Feed forward, perceptron, backpropagation trials

JANN-0.2

  • Many network structure trials
  • Transfere Functions:
  • COMPET
  • HARDLIM
  • HARDLIMS
  • LOGSIG
  • LOGSIGDIFF
  • POSLIN
  • PURELIN
  • SATLIN
  • SATLINS
  • TANSIG
  • TANSIGDIFF
  • Performance Functions:
  • MSE, MAE, SSE
Something went wrong with that request. Please try again.