Personalized Classifiers from Ensemble learning with Gaussian Process

Personalized Federated Learning with Gaussian Processes
Idan Achituve, Aviv Shamsian, Aviv Navon, Gal Chechik, Ethan Fetaya

This is a theoretical work for personalized learning with limited data. It was shown that the disadvantage of restricted exposure for each “person” or client can be remediated by learning a shared kernel function across all clients. This is then parameterized by a neural network as a personal classifier for each client.
A technical advance that is a good metaphor for the value of social interaction (including language) across individual brains. The shared kernel develops in the sphere of social interaction. Then classification in spite of limited data for each individual can be highly successful. A deep problem for any discussion of the social brain.