ncku-talk2013-12-17

NCTS(South) Seminar

DATE 2013-12-17　14:10-15:00

PLACE R204, 2F, NCTS, NCKU

SPEAKER Prof. Ilker Birble（MSIE Program, Sabanci University）

TITLE Parallel-by-Design: A New Algorithm for Unconstrained Optimization

ABSTRACT We design a parallel unconstrained optimization algorithm from the ground up. Our design efforts are centered around the idea of using multiple points at each iteration to build a representative model of the objective function. Using the new information gathered from those multiple points, a local step is gradually improved by updating its direction as well as its length. The required computations are mostly independent and separable in problem domain. Furthermore, the cost of synchronization per inner iteration is negligible. Thus, the algorithm scales well on a shared-memory system. We give a global convergence result and also provide parallel implementation details accompanied with numerical results. These results show the scalability of the proposed algorithm, its efficient resource usage and its success in improving the initial local step.

NCTS(South) Seminar

DATE	2013-12-17　14:10-15:00

PLACE	R204, 2F, NCTS, NCKU

SPEAKER	Prof. Ilker Birble（MSIE Program, Sabanci University）

TITLE	Parallel-by-Design: A New Algorithm for Unconstrained Optimization

ABSTRACT	We design a parallel unconstrained optimization algorithm from the ground up. Our design efforts are centered around the idea of using multiple points at each iteration to build a representative model of the objective function. Using the new information gathered from those multiple points, a local step is gradually improved by updating its direction as well as its length. The required computations are mostly independent and separable in problem domain. Furthermore, the cost of synchronization per inner iteration is negligible. Thus, the algorithm scales well on a shared-memory system. We give a global convergence result and also provide parallel implementation details accompanied with numerical results. These results show the scalability of the proposed algorithm, its efficient resource usage and its success in improving the initial local step.