Learning Linear Dynamical Systems from Time Series Data

Watch a research lecture from MIT's Ankur Moitra exploring linear dynamical systems and their applications in time series data analysis. Delve into a novel algorithm based on the method of moments that operates efficiently under minimal assumptions, bridging gaps in existing approaches that only offer asymptotic guarantees or require restrictive conditions. Discover how theoretical machine learning tools, particularly tensor methods, can be applied to non-stationary settings. The presentation, part of the Joint IFML/MPG Symposium at the Simons Institute, showcases collaborative research with Ainesh Bakshi, Allen Liu, and Morris Yau, demonstrating renewed interest in these systems due to their connections with recurrent neural networks.