Learning Linear Dynamical Systems via Spectral Filtering
Elad Hazan, Karan Singh, Cyril Zhang
–Neural Information Processing Systems
We present an efficient and practical algorithm for the online prediction of discrete-time linear dynamical systems with a symmetric transition matrix. We circumvent the non-convex optimization problem using improper learning: carefully overparameterize the class of LDSs by a polylogarithmic factor, in exchange for convexity of the loss functions. From this arises a polynomial-time algorithm with a near-optimal regret guarantee, with an analogous sample complexity bound for agnostic learning. Our algorithm is based on a novel filtering technique, which may be of independent interest: we convolve the time series with the eigenvectors of a certain Hankel matrix.
Neural Information Processing Systems
Oct-2-2024, 19:56:54 GMT
- Country:
- North America
- Canada > Ontario
- Toronto (0.14)
- United States
- California > Los Angeles County
- Long Beach (0.04)
- New Jersey > Mercer County
- Princeton (0.04)
- California > Los Angeles County
- Canada > Ontario
- North America
- Technology: