|
Abstract:
|
The study of dynamical systems with linear dynamics and quadratic costs is a classical topic in control theory. However, finite time regret bounds for these systems have only been derived recently. Unfortunately, existing regret bounds suffer from a number of deficiencies including stringent assumptions on the underlying system and undesirable (e.g., exponential) dependence on the dimension of the state space. In this talk, I will describe our recent efforts to mitigate these deficiencies. On the way to regret bounds, I will also touch upon finite time estimation error bounds for general (i.e., not necessarily stable) vector autoregressive (VAR) time series models.
|
