US 11,966,851 B2
Construction of a machine learning model
Aoun Lutfi, Dubai (AE)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Apr. 2, 2019, as Appl. No. 16/372,474.
Prior Publication US 2020/0320404 A1, Oct. 8, 2020
Int. Cl. G06N 20/00 (2019.01); G06N 3/084 (2023.01); G06N 3/126 (2023.01); G06N 20/20 (2019.01)
CPC G06N 3/126 (2013.01) 20 Claims
OG exemplary drawing
 
1. A computer implemented method, the method comprising:
obtaining, by at least one computer processor, a plurality of stored machine learning models;
extracting, by the at least one computer processor, a plurality of updated parameters sets from the plurality of stored machine learning models, wherein the plurality of updated parameter sets includes a standard deviation weight parameter and a standard deviation bias parameter;
representing, by the at least one computer processor, individual parameter values as parameter distributions, wherein the parameter distribution is based on a mean of the individual parameter values and a standard deviation of the individual parameter values;
creating, by the at least one computer processor, a new machine learning model based on a respective distribution of each parameter included in the plurality of updated parameters sets;
protecting, by the at least one computer processor, the new machine learning model from reverse engineering by basing the new machine learning model on the respective distribution of each parameter to protect each parameter included in the plurality of updated parameter sets from extraction; and
processing, by the at least one computer processor, at least one new query using the new machine learning model.