US 12,033,039 B2
System and method for maintaining network integrity for incrementally training machine learning models at edge devices of a peer to peer network
Subash Sundaresan, Fremont, CA (US)
Assigned to Subash Sundaresan, Fremont, CA (US)
Filed by swarmin.ai, Fremont, CA (US)
Filed on Dec. 6, 2020, as Appl. No. 17/113,070.
Claims priority of provisional application 62/978,280, filed on Feb. 18, 2020.
Prior Publication US 2021/0256421 A1, Aug. 19, 2021
Int. Cl. G06N 20/00 (2019.01); H04L 9/32 (2006.01); H04L 67/1087 (2022.01); G06F 18/21 (2023.01); G06F 18/214 (2023.01)
CPC G06N 20/00 (2019.01) [H04L 67/1093 (2013.01); G06F 18/2148 (2023.01); G06F 18/217 (2023.01); H04L 9/3213 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A certifying node for maintaining a network integrity for incrementally training machine learning models at one or more edge devices in a peer to peer network, wherein the certifying node is communicatively connected with the one or more edge devices in the peer to peer network, wherein said certifying node is configured to:
receive an incrementally updated version of a machine learning (ML) model from a first registered edge device of a plurality of registered edge devices;
quantify an amount of contribution to the incrementally updated version of the ML model by said first registered edge to compare said amount of contribution with a predetermined threshold amount of contribution;
perform at least one of (a) rejecting said incrementally updated version of said ML model from said first registered edge device if said amount of contribution is above said predetermined threshold amount of contribution, or (b) accepting said incrementally updated version of said ML model from said first registered edge device if said amount of contribution is below said predetermined threshold amount of contribution by:
verifying that each of at least one data packet associated with said incrementally updated ML model that is received from said plurality of registered edge devices,
verifying an encrypted data that specifies a base model version of said ML model from which said incrementally updated version of said ML model is derived, and
verifying that a data item used for said incrementally updated version of said ML model is not used previously by any of said plurality of registered edge devices for incremental training of said ML model to obtain a certified version of said ML model, and
transmit said certified version of said ML model to said plurality of registered edge devices for incrementally training machine learning models at said plurality of registered edge devices in the peer to peer network.