Pattern search is a derivative-free optimization method that is widely applicable to science and engineering optimization. We have extended the parallel version of pattern search to be both asynchronous and fault-tolerant, resulting in the Asynchronous Parallel Pattern Search (APPS) algorithm. This method has proven to be extremely effective on several engineering problems at Sandia National Labs and elsewhere. We will discuss issues of the parallelization, the convergence analysis, and future challenges.
Sponsored jointly with the Department of Applied Mathematics.
Hosted by Richard Byrd.
Refreshments will be served immediately following the talk in ECOT 831.