Â鶹ӰԺ

Michelle Girvan, Department of Physics, University of Maryland

Tailored Forecasts from Short Time Series Using Meta-Learning and Reservoir Computing

Machine learning (ML) models can be effective for forecasting the dynamics of unknown systems from time-series data, but they often require large datasets and struggle to generalize—that is, they fail when applied to systems with dynamics different from those seen during training. Combined, these challenges make forecasting from short time series particularly difficult. To address this, we introduce Meta-learning for Tailored Forecasting from Related Time Series (METAFORS), which supplements limited data from the system of interest with longer time series from systems that are suspected to be related. By leveraging a library of models trained on these potentially related systems, METAFORS builds tailored models to forecast system evolution with limited data. Using a reservoir computing implementation and testing on simulated chaotic systems, we demonstrate METAFORS’ ability to predict both short-term dynamics and long-term statistics, even when test and related systems exhibit significantly different behaviors, highlighting its strengthsÌý in data-limited scenarios.

​â¶Ä‹â¶Ä‹â¶Ä‹â¶Ä‹â¶Ä‹â¶Ä‹