US 12,093,400 B1
Systems and methods for model security in distributed model training applications
David Esra, San Antonio, TX (US); and Susanna Cox, Chattanooga, TN (US)
Assigned to BOBI, INC., San Antonio, TX (US)
Filed by BOBI, INC., San Antonio, TX (US)
Filed on Feb. 27, 2024, as Appl. No. 18/588,781.
Claims priority of provisional application 63/450,541, filed on Mar. 7, 2023.
Int. Cl. G06F 21/56 (2013.01); G06F 21/57 (2013.01); G06N 5/04 (2023.01); G06N 20/00 (2019.01)
CPC G06F 21/577 (2013.01) [G06F 21/56 (2013.01); G06N 5/04 (2013.01); G06N 20/00 (2019.01); G06F 2221/034 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer implemented method for edge device level security analytics, the computer implemented method comprising:
creating a first local model on a first user device by deploying a first central model to the first user device, wherein the first central model is trained to analyze user data and provide at least one predictive inference;
obtaining, in near real-time, digital data associated with a user, the digital data obtained at the first user device, the digital data associated with at least one of user physiology data and health data;
generating, at the first user device, a first data set from at least a portion of the obtained digital data;
applying, at the first user device, a local data security review to the first data set, wherein the local data security review comprises monitoring the first data set, at the first user device through an edge sensor, where the edge sensor uses at least one generated edge threshold to identify data anomalies;
filtering, at the first user device, the first data set to exclude the data anomalies;
generating, at the first user device, a second data set by applying differential privacy techniques to the first data set when no data anomalies are identified or to the filtered first data set when data anomalies are identified;
training, at the first user device, a second local model using the second data set;
applying, at the first user device, a local model security review to the second local model;
applying, at a central model processing system, a central model security review to the second local model;
combining, at the central model processing system, the second local model with a third local model received from a second user device;
auditing, at the central model processing system, performance of the first central model based on analysis of the combined second and third local models to generate audit metrics;
updating, at the central model processing system, the central model based on the audit metrics; and
deploying the updated central model to a plurality of user devices.