SNARLI is a package of Java classes for building different kinds of neural networks with a minimum of effort. It is being developed by Simon D. Levy at the Computer Science Department of Washington and Lee University.
SNARLI is graciously hosted by Sourceforge. You can directly access the following resources from here:
If you are new to SNARLI, please click here for a brief overview of its main features.
You can browse the SNARLI documentation here; it was created with the wonderful Javadoc tool.
SNARLI uses JLinAlg as its linear algebra kernel. JLinAlg is built on a solid abstract-algebra foundation and hence is of both theoretical and practical interest. I encourage everyone to take a look at it and consider signing on as a developer.
As for other neural-net packages, there are probably hundreds out there. Two large, C-based packages with GUI's that I have heard of people using successfully are SNNS and LENS. There is also an excellent summary of competitive learning methods (like the SOM and IGG classes in SNARLI) here. And though it's not a neural-net project, I have used XSB to write Definite Clause Grammars to generate training data for SNARL-based recurrent networks. Finding XSB on Sourceforge inspired me to put SNARLI up there (and steal some of their home-page style).