Improved sample complexity estimates for statistical learning control of uncertain systems

sRecently, probabilistic methods and statistical learning theory have been shown to provide approximate solutions to "difficult" control problems. Unf ortunately, the number of samples required in order to guarantee stringent performance levels may he prohibitively large. This paper Introduces bootst rap learning methods and the concept of stopping times to drastically reduc e the bound on the number of samples required to achieve a performance leve l. We then apply these results to obtain more efficient algorithms which pr obabilistically guarantee stability and robustness levels when designing co ntrollers for uncertain systems.