January 26th, 2010
The new release that we’re currently working on, will bring several very important improvments for MultiLayerPerceptron and Backpropagation. We’ve tested few networks and find out that MLP’s with Linear transfer functions in input layer learn much faster! Currently all layers including input have Sigmoid/Tanh transfer function.
The new network will have Linear transfer functions in input layer by default.
Also some users asked about Bias in MLP, and after a little research we found out that it is very important to have bias since it provides more flexibility when dividing input space, which may also be important for some problems and proviude faster learning.
Maybe the most important, we’re working on dynamic adjucting of learning rate and momentum, which will significantly speed up learning, and make it easier to use.
Also did some thread synchroniztion in easyNeurons, which resulted in few times faster and smoother training.
And thats not all. Many new handy features suggested by Neuroph community are comming!
Posted in Development | No Comments »
January 26th, 2010
Third part of the interview titled ‘Neuroph: Smart Java Apps with Neural Networks‘ has been published on developers social network NetBeans Zone. Migration of Neuroph’s GUI to NetBeans platform in cooperation with NetBeans community announced!
Posted in Project | No Comments »
December 24th, 2009
We have published the preview release of NEAT support for Neuroph. NEAT stands for “Neural Evolution of Augmeted Topologies”, it is a Genetic Algorithm for the evolution of Artificial Neural Networks (ANN) developed by Ken Stanley.
The NEAT support for Neuroph is developed and contributed by Aidan Morgan.
NEAT preview is available for download here
Other relevant links:
http://nn.cs.utexas.edu/?neat
http://en.wikipedia.org/wiki/Neuroevolution_of_augmenting_topologies
Posted in Development | No Comments »
December 13th, 2009
We’ve published basic tutorial on how to use Neuroph neural networks for prediction or function approximation:
http://neuroph.sourceforge.net/TimeSeriesPredictionTutorial.html
Posted in Uncategorized | No Comments »
December 4th, 2009
Version 2.3.1 is released publicly! This version brings important bugfixes for image recognition and several GUI improvements:
- fixed issue with editing gui in NetBeans (fixed NetBeans project file)
- fixed LMS formula
- fixed testing in black and white mode for image recognition
- fixed gui bug – exceptions when creating large networks
- changed image recognition API, so the color mode is automaticaly detected from settings used for network training removed unnecessary methods
- graph view – migrated graph view to JUNG to 2.0, created specific network layouts and removed unnecessary options
- ANT build file is now included in release which can build the jars for library and GUI.
» Download this version here
Posted in Project | No Comments »
December 4th, 2009
The Neuroph and SpaceNet (http://automenta.com/spacenet) projects are collaborating on a new kind of 3D environment for visualizing neural systems. Developers of the project share a vision of an intuitive tool for evolving and exploring the possibilities of neural networks, so future work may also include environment for 3D editing and interaction with neural systems. The developers are excited about the new features being designed.
The common software component responsible for the (easy) integration is the JUNG Graph API (http://jung.sourceforge.net).
Demo video available at: http://www.youtube.com/watch?v=c7cGSb0ITQ4
Posted in Project | No Comments »
November 25th, 2009
We’ve opened the irc chennel for support and development discussions:
Server: irc.freenode.net
Channel: #neuroph
Posted in Uncategorized | No Comments »
November 25th, 2009
An interesting integration of Neuroph with Spacenet has been made. Check out the video http://www.youtube.com/watch?v=c7cGSb0ITQ4
This could leed to 3D Neural network editor/explorer/visualizer.
Posted in Uncategorized | No Comments »
November 25th, 2009
We’re preparing the release 2.3.1 which will bring important bugfixes and improvments. The new release will fix few minor image recognition bugs, and the GUI bug when creating large networks with 1000’s of neurons. Graph view is implemented with JUNG 2.0 which provides cleaner API and better integration. Now graph view supports new network specific layouts. We’re currently testing the new relase and it will be published soon. Thanks to all users who reported bugs and helped us to improve Neuroph.
Posted in Development | No Comments »
October 27th, 2009
We have released a few patches for bugs reported for a release 2.3
The NetBeans project wth all patches is available for download here:
https://sourceforge.net/projects/neuroph/files/neuroph%202.3/neuroph_2.3.1_nb-RC1.zip/download
Patches include:
- Fix for project.xml to resolve issue with errors reported after editing GUI
- Fix for image recognition tool – testing black and white images
- Fix for LMS learning rule (small tweak)
Posted in Development | No Comments »