Abstract
We present a model-agnostic federated learning method that mirrors theoperation of a smart power grid: diverse local models, like energy prosumers,train independently on their own data while exchanging lightweight signals tocoordinate with statistically similar peers. This coordination is governed by agraph-based regularizer that encourages connected models to produce similarpredictions on a shared, public unlabeled dataset. The resulting method is aflexible instance of regularized empirical risk minimization and supports awide variety of local models - both parametric and non-parametric - providedthey can be trained via regularized loss minimization. Such training is readilysupported by standard ML libraries including scikit-learn, Keras, and PyTorch.