Generalized Connectionist Associative Memory

This paper presents a generalized associative memory model, which stores a collection of tuples whose components are sets rather than scalars. It is shown that all library patterns are stored stably. On the other hand spurious memories may develop. Applications of this model to storage and retrieval of naturallyarising generalized sequences in bioinformatics are presented. The model is shown to work well for detection of novel generalized sequences against a large database of stored sequences, and for removal of noisy black pixels in a probe image against a very large set of stored images.