ARTMAP Module

ARTMAP Neural Networks:

Inspired by biological neural networks, ARTMAP systems learn to associate arbitrary sequences of input/output pattern pairs
With fast learning, ARTMAP can encode an input in a single presentation. Many alternative methods require thousands of training epochs.
Fast learning lets ARTMAP networks do incremental learning, allowing an agent to interact with its environment in real time.

API Docs for ARTMAP
The documentation for the programmer's interface for the ARTMAP module is available here.

Source code for ARTMAP
The source code for the ARTMAP module is available in Zipped format here.

For a Java port of this code, see here.

ARTMAP DLL
If you just want to use the implementation, and don't need the source, you can use the ARTMAP implementation as a Dynamically Loadable Library (DLL).

A zip file containing the ARTMAP DLL is available here. It also contains the file ArtmapDll.lib, which is required for linking to the DLL. Also included is a simple example showing how to call the DLL: ArtmapDllTester.cpp, and a Visual C++ project file showing how to link to the DLL.

Versions of ARTMAP Provided

The module provided with the Classer toolkit implements four versions of the ARTMAP neural network:

Fuzzy ARTMAP
- Learns associations, and makes single-label predictions for arbitrary input samples
- See Fuzzy ARTMAP: A neural network architecture for incremental supervised learning of analog multidimensional maps. Carpenter, G.A., Grossberg, S., Markuzon, N., Reynolds, J.H., & Rosen, D.B. (1992). IEEE Transactions on Neural Networks, 3, 698-713.
Default ARTMAP
- The same as Fuzzy ARTMAP in training, but can make distributed predictions (in the form of a probability distributions over the set of labels)
- See Default ARTMAP. Carpenter, G.A. (2003). Proceedings of the International Joint Conference on Neural Networks (IJCNN'03), Portland, Oregon.
Instance-Counting ARTMAP
- Same as the Default model, but predictions are weighted by the "experience" of the category nodes making them
- See ARTMAP-IC and medical diagnosis: Instance counting and inconsistent cases. Carpenter, G.A. & Markuzon, N. (1998). Neural Networks, 11, 323-336.
Distributed ARTMAP
- Diverges from the other models in that training as well as testing is distributed
- See Distributed ARTMAP: A neural network for fast distributed supervised learning. Carpenter, G.A., Milenova, B., & Noeske, B. (1998). Neural Networks, 11, 793-813.

The version that is used can be specified as a parameter via the setNetworkType() method call.

Classer Toolkit Menu

ARTMAP Module