Stateful training of machine learning models in wireless networking

Abstract

A user equipment device is provided that includes at least one processor and at least one memory storing instructions. The instructions, when executed by the at least one processor, cause the user equipment device to at least: access an available machine learning (ML) model; receiving a freeze-adaptive scale value from the network device; determining a frozen portion of an untrained current ML model and an adaptive portion of a trained available ML model based on the freeze-adaptive scale value; accessing performance measurements for the available ML model; retraining an adaptive portion of the available ML model to provide a retrained ML model; determining a performance measurement for the retrained ML model; and selecting one of the retrained ML model or the available ML model based on the performance measurement of the available ML model and the performance measurement of the retrained ML model.

Images