Hierarchy Extraction

Knowledge Discovery: Self-Organized Rule Extraction
An add-on for the Classer toolkit implements the ARTMAP Information Fusion System, the basis for the a self-organized approach for extracting rules describing relationships between class labels that describe a domain of interest. More basically, the cartoon shown below illustrates the issue; in the real world, we constantly have to deal with multiple labels for objects. Rather than get confused by the apparent contradictions, the ARTMAP Information Fusion System uses these multiple labels to infer relationships. This approach is best described by the abstract for Carpenter, Martens & Ogas (2005):

    Classifying novel terrain or objects from sparse, complex data may require the resolution of conflicting information from sensors working at different times, locations, and scales, and from sources with different goals and situations. Information fusion methods can help resolve inconsistencies, as when evidence variously suggests that the class of an object is car, truck, or airplane. The methods described here address a complementary problem, supposing that information from sensors and experts is reliable though inconsistent, as when evidence suggests that the class of an object is car, vehicle, and man-made. Underlying relationships among classes are assumed to be unknown to the automated system or the human user. The ARTMAP information fusion system uses distributed code representations that exploit the capacity of the neural network for one-to-many learning in order to produce self-organizing expert systems that discover hierarchical knowledge structures. The fusion system infers multi-level relationships among groups of output classes, without any supervised labeling of these relationships. The procedure is illustrated with two image examples, but is not limited to the image domain.

The papers that make use of the ARTMAP Information Fusion System (AIFS) are listed below. Note that the Knowledge Discovery module is an add-on to Classer in the following sense. The commands that implement it are part of ClasserScript, the batch interface to Classer, but they are not part of the Classer core API. In other words, to use the AIFS, one either has to use ClasserScript, or make calls to the AIFS API, which is separate from the Classer API.

Scripts Cookbook:
A set of scripts duplicating the Boston Testbed results from Carpenter, Martens, Ogas (2005) are available here. The user should first read the file "readme.txt", which discusses the installation of some required MATLAB libraries, and then run the top-level script "doAll.csd", which in turn calls the others. This is a larger-scale set of Classer scripts, and running the full set should take 20-60 minutes, depending on the speed of your computer. In addition, the user should download the MATLAB-related support files here.

Carpenter, G.A., Martens, S., Ogas, O.J. (2004) Self-organizing hierarchical knowledge discovery by an ARTMAP image fusion system. Proceedings of the Seventh International Conference on Information Fusion (Stockholm, June 04) pp. 235-242. Technical Report CAS/CNS TR-04-001, Boston, MA: Boston University.

Carpenter, G.A., Martens, S., Mingolla, E., Ogas, O.J., Sai, C. (2004) Biologically Inspired Approaches to Automated Feature Extraction and Target Recognition. Proceedings of the 33rd Workshop on Applied Imagery Pattern Recognition -- AIPR 2004, October 13-15, 2004, Washington, DC, pp. 61-65. Piscataway, NJ:  IEEE. Technical Report CAS/CNS TR-2004-008, Boston, MA: Boston University

Carpenter, G.A., & Martens, S. (2005) Self-organizing hierarchical knowledge discovery by an ARTMAP information fusion system. Submitted to Proceedings of the International Joint Conference on Neural Networks (IJCNN'05), Montreal. Technical Report CAS/CNS TR-2005-002, Boston, MA: Boston University.

Carpenter, G.A., Martens, S., & Ogas, O.J. (2005). Self-organizing information fusion and hierarchical knowledge discovery:  a new framework using ARTMAP neural networks. Neural Networks, 18(3) pp. 287-295. Technical Report CAS/CNS TR-2004-016, Boston, MA: Boston University.