Abstract: A plurality of memory units, called "mnemonds", is each capable of storing data elements and of responding to requests to recover those data elements from storage. A plurality of mnemonds is arranged in a linear array, with each mnemond connected to be capable of responding to signals from its neighbor mnemonds. When a data structure is stored, each array stores a copy of that data structure, either allocating the elements of the data structure one per mnemond (with possible wrap-around to the beginning of the linear array) or allocating the data structures to one or a predetermined number of mnemonds. When a data structure is to be located in memory through cueing, precise correspondences between data stored and cueing data entered identify the unique location or locations of said matching data; when a data structure is to be recovered from memory, identification of its location permits recovery of data in that location or in that location and succeeding locations.

Abstract: Raw feature sensors are arranged in rank order for sensing raw features which exceed a respective set of raw feature values. Each raw feature sensor determines if a feature exists which falls within its range. The raw feature sensors have ranges which are exponentially separated by a constant factor, which may be chosen to allow easy approximate calculation of small integer ratios or other important ratios. A contrast-enhancer may be used in the case of visual data to improve the operation of the raw feature sensors. A ratio-detector network is repsonsive to the raw feature sensors and calculates ratios of the raw feature data. The ratio-detector network pairwise ANDs each possible pair of results from the raw feature sensors and combines these pairwise AND results with OR gates to generate a set of outputs, each of which indicates the detection of a single ratio.