US 12,079,732 B2
Cross-organization continuous update of edge-side event detection models in warehouse environments via federated learning
Vinicius Michel Gottin, Rio de Janeiro (BR); Pablo Nascimento da Silva, Rio de Janeiro (BR); and Paulo Abelha Ferreira, Rio de Janeiro (BR)
Assigned to DELL PRODUCTS L.P., Round Rock, TX (US)
Filed by Dell Products L.P., Round Rock, TX (US)
Filed on Mar. 14, 2022, as Appl. No. 17/694,173.
Prior Publication US 2023/0289169 A1, Sep. 14, 2023
Int. Cl. G06F 11/36 (2006.01); G06F 8/65 (2018.01); G06F 21/60 (2013.01); G06N 3/098 (2023.01); G06N 20/00 (2019.01)
CPC G06N 3/098 (2023.01) [G06F 8/65 (2013.01); G06F 11/3692 (2013.01); G06F 21/60 (2013.01); G06N 20/00 (2019.01)] 18 Claims
OG exemplary drawing
 
1. A method, comprising:
deploying, from a central node, respective instances of an event detection model to each edge node in a group of edge nodes;
providing training data to the edge nodes, wherein the training data is usable by each of the edge nodes to train its respective instance of the event detection model, and wherein the training data comprises ground truth data indicating one or more events that have actually occurred, and/or the training data comprises data obtained from an environment other than an environment in which the edge nodes operate;
receiving, by the central node from the edge nodes, gradients that capture differences between the instance of the event detection model, and updated instances of the event detection model that were updated by the edge nodes as part of a training process performed at the edge nodes;
creating a new federation by splitting an existing federation of the group of edge nodes based on drifts among the gradients, wherein each federation has its own event detection model and any two federations have different event detection models from each other;
updating, by the central node, an event detection model of each federation with the gradients to create an updated model; and
deploying, by the central node, respective instances of the updated model to each federation, to which a respective portion of the edge nodes belongs.