Pre-Grant Publication Number: 20080016013
Filing Date: July 12, 2006
Inventors: Rajesh Venkat Subbu, Stefano Romoli Bonissone
Assignee: General Electric Company
Current U.S. Classification: 706, 706/013000
View Prior Art for Claim 00012
A method for implementing a multi objective evolutionary algorithm (MOEA) on a programmable hardware device, the method comprising:
generating a sequence of pseudo random numbers and generating a population of solutions based on the sequence of pseudo random numbers;
adapting the population of solutions to generate an adapted population of solutions;
evaluating each member of the population of solutions and the adapted population of solutions, wherein the evaluation is performed using a fitness evaluator implemented on the programmable hardware device;
selecting a subset of members from the population of solutions and the adapted population of solutions to generate a filtered population of solutions, wherein the steps of evaluating and selecting are iteratively performed until one or more termination criteria are reached; and
archiving the populations of solutions.
Title Computational Intelligence and Multimedia Applications, 1999. ICCIMA apos;99. Proceedings. Third Int
ISBN
Description
In claim 1, the inventors claim a particular decomposition of a MOEA that mirrors the generally understood structure of the algorithm. Object-oriented software systems typically use the same or very similar decompositions. The decomposition described in this paper is identical except that it groups the fitness evaluator and the dominance filter into one component (see figure 1). The random number generator is included by not illustrated. The paper also allows archiving of fitness evaluations according to user specifications.
Title Genetic Programming and Evolvable Machines, Volume 2
ISBN
Description
This paper is one of many (do a Google on "fpga fitness function" for more) that describes the use of a programmable hardware device (fpga) for fitness evaluation in a genetic algorithm.
