Parallel MOEA/D-ACO on GPU

Abstract

This paper describes the idea of MOEA/D-ACO (Multiobjective Evolutionary Algorithm based on Decomposition and Ant Colony Optimization) and proposes a Graphics Processing Unit (GPU) implementation of MOEA/D-ACO using NVIDIA CUDA (Compute Unified Device Architecture) in order to improve the execution time. ACO is well-suited to GPU implementation, and both the solution construction and pheromone update phase are implemented using a data parallel approach. The parallel implementation is applied on the Multiobjective 0-1 Knapsack Problem and the Multiobjective Traveling Salesman Problem and reports speedups up to 19x and 11x respectively from the sequential counterpart with similar quality results. Moreover, the results show that the size of test instances, the number of objectives and the number of subproblems directly affect the speedup.