SCN Artificial Neural Network Library
SCN Artificial Neural Network Library provides a programmer with a simple object oriented API for constructing ANNs. It supports non-recursive networks with an arbitrary number of layers, each with an arbitrary number of nodes.
| Tags | Scientific/Engineering Artificial Intelligence |
|---|---|
| Licenses | LGPL |
| Implementation | C++ |
Recent releases
- 27 Sep 2001 07:06
No changes have been submitted for this release.
- All comments
Recent commentsNOTICE
Development of this project has been discontinued.
There are plenty of other GPL-ed neural network libraries out there
(Amygdala (http://amygdala.sourceforge.net/),
annie (http://annie.sourceforge.net/),
libperceptronnetwork (http://user.cs.tu-berlin.de/~nowozin/libpn/),
libneural (http://ieee.uow.edu.au/~daniel/software/libneural/)), and
I felt that building another was not necessary. Downloads of this
library will be available, probably indefinately, but there are no plans
to release new versions at this point. If someone is interested in
picking up the project and improving upon it, I would love to hear about
it. I would be willing to provide hosting resources, in this case.
New Development Mailing List
Well, I finally found time to get this set up. If you are interested in helping with the project, or just want to find out what the plans are for future releases, subscribe by sending a message to scnannlib-dev-subscribe@sentinelchicken.org.
Warning: Drastic Changes to API
Since the first release, I have received some very good feedback in how the library could be improved. To accommodate many of these improvements, there will be massive changes to the API in the next release. Therefore, be a little cautious in spending too much time writing code that uses the current library API.
Design is still in the works, so if there are any suggestions on how it could be done, by all means let me know. So far the plan is to generalize and simplify the network interface and use a seperate factory class to create the networks. In this way, we can create many different network classes that are all generated through the factory. Each network class could then be very specific in its speed/memory optimizations.
Comments either by email or posting are welcome.
Bug info
There has been a bug found in one of the testing scripts for the library. This caused the process on a Red Hat machine to segfault after exactly 15 trainings. The code has been fixed in the repository, but I hadn't planned on releasing it right now, since the changes are minor, and shouldn't affect users who are writing their own programs who use only the library. However if there is significant demand for this bugfix, then I will release it. Please write me if you have any questions/comments.