Logarithmic Regret for Learning Linear Quadratic Regulators Efficiently

We consider the problem of learning in Linear Quadratic Control systems whose transition parameters are initially unknown. Recent results in this setting have demonstrated efficient learning algorithms with regret growing with the square root of the number of decision steps. We present new efficient...

Descripción completa

Detalles Bibliográficos

Autores principales:	Cassel, Asaf, Cohen, Alon, Koren, Tomer
Formato:	Artículo
Publicado:	2020
Materias:	Machine Learning
Acceso en línea:	https://arxiv.org/abs/2002.08095 http://arxiv.org/abs/2002.08095
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!