A parallel evolutionary algorithm for unconstrained binary quadratic problems

In this paper an island model is described for the unconstrained Binary Quadratic Problem (BQP), which can be used with up to 2500 binary variables. Our island model uses a master-slave structure and the migration is centralized. In the model a basic evolutionary algorithm (EA) runs which is a hybrid, steady-state EA. The basic EA uses a new mutation operator that is composed of two parts and based on a modified version of an explicit collective memory method (EC-memory), the Virtual Loser [2].We tested our island model on the benchmark problems from the OR-Library. Comparing the results with other heuristic methods, we can conclude that our algorithm is highly effective in solving large instances of the BQP; it has a high probability of finding the best-known solutions. Categories and Subject Descriptors 12.8 [Artificial Intelligence]: Problem Solving, Control Methods and Search – heuristic methods General Terms Algorithms. Keywords Binary quadratic programming; evolutionary algori...

• Type: Conference
• Year: 2008
• Where: GECCO
• Authors: István Borgulya
• Keywords: Binary Quadratic Problem, Evolutionary Algorithm, GECCO 2008, Island Model, Optimization