Certainty equivalence control with forcing: revisited

摘要

Summary form only given, as follows. Stochastic adaptiveoptimization problems are considered with the objective of minimizingthe rate of increase of the learning loss, i.e. the additional cost onehas to pay due to the inbuilt learning tasks in such problems. Inparticular, an examination is made of two problems: the multiarmedbandit problem, and the adaptive control of Markov chains. Previous workhas shown that the minimum rate of increase of the learning loss forthese problems is typically