US 12,147,879 B2
Federated learning with dataset sketch commitment based malicious participant identification
Lei Yu, Sleepy Hollow, NY (US); Qi Zhang, West Harrison, NY (US); Petr Novotny, Mount Kisco, NY (US); and Taesung Lee, Ridgefield, CT (US)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Feb. 22, 2021, as Appl. No. 17/180,972.
Prior Publication US 2022/0269977 A1, Aug. 25, 2022
Int. Cl. G06N 20/00 (2019.01)
CPC G06N 20/00 (2019.01) 20 Claims
OG exemplary drawing
 
1. A method, in a data processing system comprising at least one processor and at least one memory, the at least one memory comprising computer executable code that, when executed by the at least one processor, causes the at least one processor to be specifically configured to implement a federated machine learning update (FMLU) engine, the method comprising:
receiving, by the FMLU engine, from a plurality of participant computing systems, a plurality of machine learning (ML) computer model parameter updates for updating a federated ML computer model;
receiving, by the FMLU engine, from the plurality of participant computing systems, a plurality of dataset sketch commitment data structures, wherein each dataset sketch commitment data structure provides statistical characteristics of a corresponding local dataset used by a corresponding participant computing system to train a local ML computer model that is local to the participant computing system;
performing, by the FMLU engine, a potentially malicious participant identification operation based on an analysis of the plurality of dataset sketch commitment data structures to identify one or more potentially malicious participants based on differences in dataset sketch commitment data structures;
discarding, by the FMLU engine, ML computer model parameter updates received from participant computing systems identified as potentially malicious participants to thereby generate a modified set of updates; and
updating parameters of the federated ML computer model based on the modified set of updates.