US 12,244,622 B2
Method to detect and defend against targetted adversarial attacks on a federated learning system
Sherin Mathews, Fremont, CA (US); and Samuel Assefa, Watertown, MA (US)
Assigned to U.S. Bancorp, National Association, Minneapolis, MN (US)
Filed by U.S. Bancorp, National Association, Minneapolis, MN (US)
Filed on Dec. 12, 2022, as Appl. No. 18/079,782.
Prior Publication US 2024/0195826 A1, Jun. 13, 2024
Int. Cl. H04L 29/06 (2006.01); G06N 20/00 (2019.01); H04L 9/40 (2022.01)
CPC H04L 63/1425 (2013.01) [G06N 20/00 (2019.01)] 20 Claims
OG exemplary drawing
 
1. A method, comprising:
transmitting, by a server, a first client model to a first computing device and a second client model to a second computing device,
wherein, in a first training iteration, the first computing device trains the first client model and the second computing device trains the second client model;
determining, by the server based on the first training iteration, (i) a first predictive score ratio for the first computing device by comparing a first output of the first client model with a first output of a global model, and (ii) a first predictive score ratio for the second computing device by comparing a first output of the second client model with the first output of the global model;
determining, by the server, the first computing device and the second computing device match responsive to determining the first predictive score ratio for the first computing device and the first predictive score ratio for the second computing device are within a cluster threshold of each other,
wherein, in a second training iteration subsequent to the first training iteration, the first computing device trains the first client model and the second computing device trains the second client model;
determining, by the server based on the second training iteration, (i) a second predictive score ratio for the first computing device by comparing a second output of the first client model with a second output of the global model, and (ii) a second predictive score ratio for the second computing device by comparing a second output of the second client model with the second output of the global model; and
detecting, by the server, an anomaly in the first computing device responsive to (i) the determining the first computing device and the second computing device match, and (ii) determining the second predictive score ratio for the first computing device exceeds the second predictive score ratio for the second computing device by an amount above a difference threshold.