US 12,248,601 B2
Privacy-preserving machine learning training based on homomorphic encryption using executable file packages in an untrusted environment
Amin Hassanzadeh, Arlington, VA (US); Neil Hayden Liberman, Arlington, VA (US); Aolin Ding, Piscataway, NJ (US); and Malek Ben Salem, Falls Church, VA (US)
Assigned to Accenture Global Solutions Limited, Dublin (IE)
Filed by Accenture Global Solutions Limited, Dublin (IE)
Filed on Jul. 22, 2021, as Appl. No. 17/383,314.
Prior Publication US 2023/0025754 A1, Jan. 26, 2023
Int. Cl. G06F 21/62 (2013.01); G06F 21/60 (2013.01); G06N 20/00 (2019.01); H04L 9/00 (2022.01)
CPC G06F 21/6245 (2013.01) [G06F 21/602 (2013.01); G06N 20/00 (2019.01); H04L 9/008 (2013.01)] 11 Claims
OG exemplary drawing
 
1. A method for training machine learning models, the method comprising:
based on execution, by one or more processors, of an executable file package stored at a memory:
receiving, by the one or more processors from each of one or more client devices, encrypted data and a public key,
wherein the executable file package includes one or more configuration files, one or more machine learning (ML) libraries, and one or more homomorphic encryption (HE) libraries, and
wherein, for each client device of the one or more client devices, the encrypted data is homomorphically encrypted by the client device based on the public key;
applying, based on execution of the executable file package, one or more weights to the encrypted data prior to aggregating the encrypted data, wherein the one or more weights are based on processing resources available at a respective client device, an amount of client data associated with the respective client device, or a combination thereof,
aggregating, by the one or more processors, the encrypted data associated with each of the one or more client devices to generate aggregated data;
constructing, by the one or more processors, an ML model based on the aggregated data, the ML model configured to output a predicted diagnosis based on input data; and
initiating, by the one or more processors, deployment of the ML model to at least one of the one or more client devices, an endpoint node, or a combination thereof.