US 12,301,430 B2
Hidden-layer routing for disaggregated artificial neural networks
Robert Edgar Barton, Richmond (CA); Frank Brockners, Cologne (DE); Jerome Henry, Pittsboro, NC (US); Matthias Falkner, Ottawa (CA); Indermeet Singh Gandhi, San Jose, CA (US); and Thomas Michel-Ange Feltin, Paris (FR)
Assigned to Cisco Technology, Inc., San Jose, CA (US)
Filed by Cisco Technology, Inc., San Jose, CA (US)
Filed on Sep. 28, 2023, as Appl. No. 18/477,476.
Prior Publication US 2025/0112832 A1, Apr. 3, 2025
Int. Cl. H04L 41/16 (2022.01); H04L 45/74 (2022.01); H04L 67/1008 (2022.01)
CPC H04L 41/16 (2013.01) [H04L 45/74 (2013.01); H04L 67/1008 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method performed at a controller, comprising:
receiving, at the controller, a deep learning model;
determining, at the controller, a graph structure of the deep learning model received at the controller, wherein the graph structure includes neurons organized in layers, and the layers include:
an input layer;
a plurality of hidden layers comprising a first hidden layer and a second hidden layer;
an output layer;
obtaining, at the controller, topology data indicating a network topology of edge devices in a network, wherein the network includes:
a first Internet Protocol (IP) subnetwork (subnet) including first IP addresses that are associated with first edge devices of the edge devices; and
a second IP subnet including second IP addresses that are associated with second edge devices of the edge devices;
assigning, at the controller, a first portion of the first IP addresses of the first IP subnet to at least a part of the first hidden layer;
assigning, at the controller, a second portion of the second IP addresses of the second IP subnet to at least a part of the second hidden layer such that the first and second hidden layers at least partially execute in different subnets;
deploying the first portion of the first hidden layer to the first edge devices based at least in part on the first portion of the first hidden layer being assigned to the first portion of the first IP addresses, wherein the first edge devices collectively execute the first portion of the first hidden layer to perform first functionality of the first hidden layer; and
deploying the second portion of the second hidden layer to the second edge devices based at least in part on the second portion of the second hidden layer being assigned to the second portion of the second IP addresses, wherein the second edge devices collectively execute the second portion of the second hidden layer to perform first functionality of the second hidden layer.