NCTS(South) Seminar


DATE2011-05-26¡@16:00-17:00

PLACER204, 2F, NCTS, NCKU

SPEAKERProf. Weichung Wang ¤ý°¶¥ò±Ð±Â¡]Dept. of Math., NTU¡^

TITLEOptimizations without Derivative Information

ABSTRACT Optimization problems involving complicated or black-box objective functions arising in various applications like engineering designs, computer codes tuning, and medical imaging. Gradient based optimization schemes are not suitable in such problems due to expensive evaluations or lack of the derivatives. Instead, we can shape up efficient methods to tackle the problems by combining interdisciplinary tools including data driven meta-models (e.g. Kriging), bio-inspired stochastic schemes (e.g. particle swarm), and high-performance computing (e.g. GPU). We illustrate such ideas on an example targeting to minimize the bandgaps of three dimensional photonics crystals.