| 摘要 |
A network system analyzes input words for the search and retrieval of pertinent information. The novel system then selects a module of a self organizing map (SOM) which contains the same dimension of classification as a selected input word and where neurons are interconnected horizontally (between modules) and vertically (at input and inside a module) by arrays of link weights. The system then determines what nodes or processing units within the SOM will be activated and subsequently compared to the selected input word. Feedback is utilized via a punishment/reward scheme to adjust the link weights so that the system learns the best paths and/or methods to create acceptable decisions or outputs.
|
