no code implementations • 4 Feb 2022 • Sercan O. Arik, Nathanael C. Yoder, Tomas Pfister
Real-world time-series datasets often violate the assumptions of standard supervised learning for forecasting -- their distributions evolve over time, rendering the conventional training and model selection procedures suboptimal.