US 11,853,878 B2
Large-scale networks of growth transform neurons
Shantanu Chakrabartty, St. Louis, MO (US); and Ahana Gangopadhyay, St. Louis, MO (US)
Assigned to Washington University, St. Louis, MO (US)
Appl. No. 16/462,805
Filed by Shantanu Chakrabartty, St. Louis, MO (US); and Ahana Gangopadhyay, St. Louis, MO (US)
PCT Filed Nov. 22, 2017, PCT No. PCT/US2017/062986
§ 371(c)(1), (2) Date May 21, 2019,
PCT Pub. No. WO2018/098257, PCT Pub. Date May 31, 2018.
Claims priority of provisional application 62/484,669, filed on Apr. 12, 2017.
Claims priority of provisional application 62/425,372, filed on Nov. 22, 2016.
Prior Publication US 2019/0370653 A1, Dec. 5, 2019
Int. Cl. G06N 3/08 (2023.01); G06N 3/049 (2023.01); G06N 20/10 (2019.01)
CPC G06N 3/08 (2013.01) [G06N 3/049 (2013.01); G06N 20/10 (2019.01)] 20 Claims
OG exemplary drawing
 
1. A growth transform neural network system comprising a computing device, the computing device comprising at least one processor and a memory storing a plurality of modules, each module comprising instructions executable on the at least one processor, the plurality of modules comprising:
a growth transform neural network module to define a plurality of mirrored neuron pairs comprising a plurality of first components and a plurality of second components each mirrored neuron pair comprising one first component and one second component connected by a normalization link wherein the plurality of first components are interconnected according to an interconnection matrix and the plurality of second components are interconnected according to the interconnection matrix, wherein each of the plurality of mirrored neuron pairs is associated with internal normalized variables p+ and p− that satisfy a normalization criterion of p++p−=1;
a growth transform module to update each first component of each mirrored neuron pair of a plurality of mirrored neuron pairs according to a growth transform neuron model; and
a network convergence module to converge the plurality of mirrored neuron pairs to a steady state condition by solving a system objective function subject to at least one normalization constraint;
wherein the growth transform neuron model is mutually coupled with a network objective function of a machine learning model;
wherein each first component of each mirrored neuron pair is configured to receive a first external input and to produce a first output and each second component of each mirrored neuron pair is configured to receive a second external input and to produce a second output; and
wherein the plurality of modules form a spiking support vector machine (SVM) and are are trained to cooperatively classify data input to the spiking SVM and to output a result of the classifying.