MX429112B - Entropic associative memory system - Google Patents

Abstract

The present invention relates to an entropic associative memory system to record, recognize, and reward distributed representations of images as finite functions stored in Register Associative Memory (RMAs). It consists of tables with columns corresponding to the arguments and rows corresponding to their displayed values, associated computational entropy, determining their operating range with an auxiliary input and output register, and a table of the exact dimensions for writing and reading the content of the RMAs of a structured bus. Also, said operations correspond to the functions from the indeterminate relational computation of the inclusive disjunction and the logical material implication between the corresponding cells of the RMA and its auxiliary record and to the selection of a random value centered on the cue accessing the memory for each argument. The retrieval operation based on material implication naturally implements the direct rejection of signals not contained in a me mory, in direct opposition to associative memory models in the neural network paradigm that always return the object most similar to implicitly stored, performing memory operations in parallel in the hardware device and integrated circuit embodiments.