US 11,893,456 B2
Device type classification using metric learning in weakly supervised settings
David Tedaldi, Zurich (CH); Pierre-Andre Savalle, Rueil-Malmaison (FR); Sharon Shoshana Wulff, Zurich (CH); Jean-Philippe Vasseur, Saint Martin d'uriage (FR); and Grégory Mermoud, Veyras VS (CH)
Assigned to CISCO TECHNOLOGY, INC., San Jose, CA (US)
Filed by Cisco Technology, Inc., San Jose, CA (US)
Filed on Jun. 7, 2019, as Appl. No. 16/434,274.
Prior Publication US 2020/0387746 A1, Dec. 10, 2020
Int. Cl. G06N 20/00 (2019.01); H04L 41/0893 (2022.01); G06F 18/23 (2023.01); G06F 18/241 (2023.01)
CPC G06N 20/00 (2019.01) [G06F 18/23 (2023.01); G06F 18/241 (2023.01); H04L 41/0893 (2013.01)] 12 Claims
OG exemplary drawing
 
1. A method comprising:
receiving, at a device classification service, telemetry data indicative of behavioral characteristics of a plurality of devices in a network;
obtaining, by the service, side information for the telemetry data by:
providing, to a user interface, the telemetry data for a pair of the plurality of devices selected by the service based on a measure of pairwise similarity between their behavioral characteristics, and providing, to the use interface, an option for a user to reject the pair selected based on the measure of pairwise similarity as dissimilar, wherein the side information comprises an indication from the user interface as to whether the user considers the pair to be similar or dissimilar;
wherein the service selects the pair of devices based on the telemetry data for the pair of devices being close to each other in a feature space and the pair of devices not belonging to the same device cluster;
applying, by the service, metric learning to the telemetry data and side information, to construct a distance function;
using, by the service, the distance function to cluster the telemetry data into device clusters; and
associating, by the service, a device type label with a particular device cluster,
wherein the service selects the pair of devices based on the telemetry data for the pair of devices being close to each other in a feature space and the pair of devices not belonging to the same device cluster.