pipette
ENEnglish

Learning structured linear dynamical systems from missing observations

Aravinda Kanchana Ruwanpathirana, Hemant Tyagi, Sunny G.W. Wang

Preprint

En palabras de los autores

We consider the problem of learning structured linear dynamical systems over convex sets , where only a small subset of the observations are available at each time point. An estimator which minimizes a bias-corrected, potentially non-convex objective function is proposed. Non-asymptotic bounds are obtained for the statistical error, which depend on the local complexity of , the trajectory length , and the sub-sampling probability . Convergence of the projected gradient descent algorithm is also established. The general theory is applied to settings where (i) is a subspace, (ii) is the set of bi-isotonic matrices, and (iii) is the set of matrices whose rows are formed by sampling Lipschitz functions. We show meaningful recovery of the transition matrix is possible for values of much smaller than what is required in the unconstrained case, and for .

Resultado principalEl resumen no menciona limitaciones.

Apareció: viernes, 9 de octubre. arXiv. Preprint, todavía sin revisión por pares.

Comentario de los autores: 62 pages, 3 figures