Supervised Federated Neural Architecture Search and Its Application in Power System Forecasting

This paper investigates the merits and limitations of various neural architecture search (NAS) methods and proposes a novel method of engaging two or more NAS methods and improving their individual performances by introducing a supervising agent. Empirical results show that the proposed method achieves a lower root mean square error (RMSE) in time-series forecasting tasks in power systems and a lower time of convergence compared to baseline NAS methods.