The first version of AI-NNFlex is available from CPAN. It's designed to be
a very flexible way of creating neural nets. The basic API creates a mesh
of nodes with properties (activation, error, activation function,
connected nodes etc). Activation flow is controlled by a plugin - at the
moment there is only a feedforward plugin. Learning is handled by another
plugin - at the moment there is a backprop plugin, and a simple
'reinforcement' plugin that just strengthens active links.
Its not especially quick - I sacrified speed for flexibility. If you're
interested please download and play. There are implementations of xor &
and in the examples directory.
Contact details for bug reports & feature requests are in the readme & I
check this mailing list about 1ce a week.