Secure and low-traffic federated network based identification of outliers in a population of electronic devices

The invention regards a method for identifying outliers in a population of electronic devices, the method comprising the steps of: providing a plurality of first machine learning models, each first machine learning model associated with one electronic device in a population of electronic devices, wherein each first machine learning model is incrementally trained based on estimated input power loss and/or input ambient temperature and/or output heat sink temperature and/or voltage of each electronic device; providing a second machine learning model, the second machine learning model being based on a federated average of model parameters of a plurality of, such as a majority of, or all of, the first machine learning models; and comparing the second machine learning model with each of the first machine learning models to identify outliers among the population and wherein the electronic devices are power converters or power switches or batteries or battery systems used for power applications.