| 摘要 |
The self-organizing pattern detection and matching system (100) of the present invention permits multiple associative memories (114, 118) to learn, without prior knowledge of the classes, the patterns in a sample of data. The self-organizing detection system learns the pattern by feedforward adaptation using symmetric triggering. Inputs to the pattern detection system (100) are applied directly to the associative memories (114, 118) which produce a best match pattern output (154) for each class. Each memory (114, 118) can contain data of a different logical type. A match strength signal is produced for each applied signal and each associated memory (114, 118). The match strength signals are processed to produce a logical match signal for each memory which are combined to provide logical trigger signals (128, 142) which allow acceptable patterns to be used to update each associative memory (114, 118). The architecture provides multi-map, self-organizing pattern recognition which allows concurrent massively parallel learning of features using different maps for each feature type. |
