US 12,014,263 B2
Methods and systems for encoding and processing vector-symbolic representations of continuous spaces
Aaron Russell Voelker, Stittsville (CA); Christopher David Eliasmith, Waterloo (CA); Brent Komer, London (CA); and Terrence Stewart, Waterloo (CA)
Assigned to APPLIED BRAIN RESEARCH INC., Waterloo (CA)
Filed by Applied Brain Research Inc., Waterloo (CA)
Filed on Mar. 18, 2020, as Appl. No. 16/823,245.
Claims priority of provisional application 62/820,089, filed on Mar. 18, 2019.
Prior Publication US 2020/0302281 A1, Sep. 24, 2020
Int. Cl. G06N 3/06 (2006.01); G06N 3/049 (2023.01); G06N 3/063 (2023.01)
CPC G06N 3/063 (2013.01) [G06N 3/049 (2013.01)] 8 Claims
OG exemplary drawing
 
1. A computer system for processing a high-dimensional vector representation of a data structure, for use in processing in artificial systems, the computer system comprising:
digital circuitry configured to obtain an input representation of a data structure, the data structure representing a continuous space;
the digital circuitry configured to generate a vector representation of the data structure by executing a plurality of binding subsystems that implement a fractional binding operation that generates slots and/or fillers ranging over spaces in which one or both of slots and fillers are separated by a continuous distance function,
the digital circuitry further configured to execute a plurality of unbinding subsystems that implement an approximate inverse of the fractional binding operation that disassociates slot-filler pairs ranging over spaces in which one or both of slots and fillers are separated by the continuous distance function,
wherein said plurality of binding subsystems and said plurality of unbinding subsystems are comprised in an artificial neural network implemented in network layers, and wherein each said network layer comprises a plurality of nonlinear components, and each said nonlinear component is configured to generate an output in response to said input representation, and wherein said output from each said nonlinear component is weighted by coupling weights of corresponding weighted couplings and weighted outputs are provided to coupled said network layers.