Abstract
The Entropic Associative Memory is an auto-associative memory in which objects are represented and stored as discrete functions or "traces" in a table, so the memory content is a 2D relation or "memory plane". Memory traces are "overlapped" on the medium, the memory is indeterminate, and the system is entropic. The memory retrieval operation produces an object out of a cue and the indeterminate memory mass, and the memory is constructive. In this paper, we present its extension to the hetero-associative case. Pairs of hetero-associated objects, possibly of different domain and/or modalities, are held in a 4D relation. The cue to a memory retrieval operation selects a largely indeterminate 2D hetero-associated memory plane, but there is no cue left to recover the object from such plane. We propose three incremental methods to address such missing cue problem, which we call random, sample and test, and search and test. The model is assessed with composite recollections consisting of manuscript digits and letters selected from the MNIST and EMNIST corpora, respectively, such that cue digits retrieve their associated letters and vice versa. We show the memory performance and illustrate the memory retrieval operation using all three methods. The system shows promise for storing, recognizing, and retrieving very large sets of objects with very limited computing resources. We also discuss the implications of the model for the psychology and the neuroscience of memory.