Sparse distributed memory for multivalued patterns
Jukka J. Vanhala, Jukka P. P. Saarinen, Kimmo K. Kaski · 2002
Kanerva's sparse distributed memory is developed for handling binary patterns. This algorithm is extended to be used with patterns of multivalued elements and applied to a gray-scale image recognition problem. Three different address activation methods are tested to identify the most useful methods. For data storage and retrieval, a method of minimal reading is proposed. Results concerning the recall accuracy of the three activation methods that work with multivalued data are discussed.>