The GLOBAL Optimization Algorithm

Abstract

The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use.

Towards Automated Worst-Case Analysis of Circuits: Selecting Initial Values for Global Optimization

“…Our approach of determining a set of initial values is based on function sampling. The core idea has been used for empirical estimation of function shape and properties [6], [7] as well as by global optimization methods [4], [5]. The latter sample the function in a large number of random points and select only a few of them.…”

“…In some circuit analysis problems, however, the function describing the circuit property in question can have multiple local optima, thus global optimization strategies are necessary. One approach is to sample the function and use clustering to select candidate points to be used as initial values and start a traditional local optimization algorithm from all of them [4], [5]. This way there will be at least one point from where the local optimizer can converge to the global optimum.…”

“…Further research include evaluating and comparing the performance of additional global optimization methods (e.g. GLOBAL [5] and memetic algorithms [4]).…”

ParallelGlobal with Low Thread Interactions

“…Global is a global optimizer designed to solve black box unconstrained optimization problems with a low number of function evaluations and probabilistic guarantees [1,2,3,4,6,8,9,11]. It uses local search algorithms to refine multiple sample points hence Global is a multi-start method.…”

“…In recent years Global was further developed [6] and it has several applications [5,7] where it aids mostly other research works. To speed up optimization processes we developed an algorithm [1] that is capable of utilizing multiple computational threads of a single machine. It cannot be directly implemented for distributed systems as the millisecond order of magnitude latency in communication would significantly slow down the synchronization of threads.…”