US 12,302,132 B2
Federated learning of growing neural gas models
Szabolcs Nováczki, Budapest (HU); Péter Szilágyi, Budapest (HU); and Lajos Bajzik, Budapest (HU)
Assigned to Nokia Solutions and Networks Oy, Espoo (FI)
Filed by Nokia Solutions and Networks Oy, Espoo (FI)
Filed on Apr. 23, 2024, as Appl. No. 18/643,994.
Claims priority of application No. 20235467 (FI), filed on Apr. 26, 2023.
Prior Publication US 2024/0365135 A1, Oct. 31, 2024
Int. Cl. H04W 24/02 (2009.01); H04L 43/0823 (2022.01); H04W 24/08 (2009.01)
CPC H04W 24/02 (2013.01) [H04L 43/0823 (2013.01); H04W 24/08 (2013.01)] 20 Claims
OG exemplary drawing
 
1. An apparatus comprising at least one processor, and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to:
receive a plurality of local growing neural gas models from a plurality of distributed trainers comprised in a plurality of radio access network nodes or in a plurality of multi-access edge computing servers,
wherein a local growing neural gas model of the plurality of local growing neural gas models represents a local state model of at least one radio access network node,
wherein the local growing neural gas model was trained based on local training data comprising a set of performance counters that are metrics used to measure and monitor performance and behavior of the at least one radio access network node; and
train a global growing neural gas model for anomaly detection or optimization of the plurality of radio access network nodes based on the plurality of local growing neural gas models,
wherein the global growing neural gas model represents a global state model of the plurality of radio access network nodes,
wherein the global growing neural gas model is trained by:
merging all of the plurality of local growing neural gas models into the global growing neural gas model at a same time; or
merging the plurality of local growing neural gas models one at a time into the global growing neural gas model according to a sequence in which the plurality of local growing neural gas models are received.