 previous | upcoming  |
 |
Jason M. Kinser Department of Bioinformatics & Computational Biology, George Mason University During the past few decades many researchers have proposed associative memory algorithms which have included neural networks, self-organizing maps, and filters. However, none seemed capable of solving realistic problems without heavily massaging the data before it reached the associative memory. These methods also fall short when considering a problem that contains multiple types (domains) of data, multiple measures of similarity between the data, and incomplete data samples. An example for such a problem is a forensic crime scene in which the data may consist of videos, fingerprints, DNA, etc. but any single scene may only have a subset of such data. Furthermore, the problem may require multiple similarity measures between two samples of data from the same domain (e.g., video) or similarity measures between data of different domains. The solution presented here constructs a single linear space that is capable of containing multiple data domains and representing all of the measures of similarities between them. This system is capable of performing multi-domain clustering, multi-domain memory association, and even state prediction. |