|
VDKAnn: ARTIFICIAL NEURAL
NETWORKS FOR VDK
A C++ library to design, construct and test connectionist
systems.
VDKAnn FEATURES
-
Supports several models of neural networks:
-
BAM system: bidirectional associative memories.
-
MLFN: Multi Layer Feedforward Networks.
-
PNN: Probabilistic Neural Network.
-
Kohonen maps: two layers self-organizing neural networks.
-
Several learning alghoritms and optimization tecniques are supported:
-
Conjugate gradient
-
Levenberg-Marquardt
-
Stochastic smoothing
-
Genetic Algorithms
-
Simulated Annealing
-
Net operations can easily embedded into C++ source code with high-level
instructions.
-
VDKAnn Requires VDK
version >= 1.2.0
-
VDKAnn is licensed under LGPL
-
VDKAnn author is Mario
Motta
FUTURE PLANS
-
A GPL'ed VDKAnn based tool to visually build and test neural networks
is under development, its name will be NetBuilder and is expected
to be released as prototype in May - June 2000
A CASE STUDY EXAMPLE
-
An example of VDKAnn used to show local minima problems in a
xor net is showed here.
|