The ARGOT strategy III: the BBN Butterfly multiprocessor

Abstract

The ARGOT strategy combines genetic algorithms with a mechanism providing a dynamically adaptive representation to form a robust optimization tool, as previously shown in the uniprocessor environment. For implementation of ARGOT on the BBN Butterfly multiprocessor, a parallel selection algorithm and a method of incremental payoff update were developed. These lead to enhanced parallelism and reduced the amount of computation needed by any genetic algorithms, including ARGOT. Experimental results on two matrix problems are presented, one a linear system from a FEM problem, and the other a nonlinear problem not well-behaved enough for consistent conjugate gradient results.<>