US Patent Application 17659146. ADAPTIVE GRADIENT COMPRESSOR FOR FEDERATED LEARNING WITH CONNECTED VEHICLES UNDER CONSTRAINED NETWORK CONDITIONS simplified abstract
ADAPTIVE GRADIENT COMPRESSOR FOR FEDERATED LEARNING WITH CONNECTED VEHICLES UNDER CONSTRAINED NETWORK CONDITIONS
Organization Name
Inventor(s)
Paulo Abelha Ferreira of Rio de Janeiro (BR)
Pablo Nascimento Da Silva of Niteroi (BR)
Roberto Nery Stelling Neto of Rio de Janeiro (BR)
Vinicius Michel Gottin of Rio de Janeiro (BR)
ADAPTIVE GRADIENT COMPRESSOR FOR FEDERATED LEARNING WITH CONNECTED VEHICLES UNDER CONSTRAINED NETWORK CONDITIONS - A simplified explanation of the abstract
- This abstract for appeared for US patent application number 17659146 Titled 'ADAPTIVE GRADIENT COMPRESSOR FOR FEDERATED LEARNING WITH CONNECTED VEHICLES UNDER CONSTRAINED NETWORK CONDITIONS'
Simplified Explanation
The abstract describes a method used by a group of edge nodes to communicate with a central node. The edge nodes generate a vector that contains gradients associated with a model instance. They then check if the model instance is overfitting to the data generated at the edge node. If overfitting is not indicated, they perform sign compression on the vector. If overfitting is indicated, they perform random perc sign compression on the vector. Finally, the compressed vector is transmitted to the central node that contains the central model.
Original Abstract Submitted
One example method includes, in an edge node, of a group of edge nodes that are each operable to communicate with a central node, performing operations that include generating a vector that includes gradients associated with a model instance, of a central model, that is operable to run at the edge node, performing a check to determine whether the model instance is overfitting to data generated at the edge node, and either performing sign compression on the vector when overfitting is not indicated, or performing random perc sign compression on the vector when overfitting is indicated, and transmitting the vector, after compression, to the central node that includes the central model.
