GPU-Accelerated Ant Colony Optimization

Publisher Summary This chapter discusses how GPGPU computing can be used to accelerate ant colony optimization (ACO) algorithms. ACO is a general term used to describe the subset of swarm intelligence algorithms that are inspired by the behaviors exhibited by colonies of real ants in nature. Current literature has shown that ACO algorithms are viable methods for tackling a wide range of hard optimization problems including the traveling salesman, quadratic assignment, and network routing problems. ACO algorithms will be seen as competitive with traditional methods for a range of problems. As a case study, the GPU-based AntMinerGPU algorithm is presented, which is an ACO algorithm for rule-based classification, and it is illustrated, how the GPGPU computing model can be leveraged to improve overall performance. Although AntMinerGPU is a special-purpose ACO algorithm, the general implementation strategy presented here is applicable to other ACO systems, swarm intelligence algorithms, and other algorithms that exhibit a similar execution pattern. In general, the accuracy of GPU- and CPU-based implementations were very similar, but in terms of running time, the GPU-based implementation is up to 100x faster than the simple CPU-based implementation for large populations of ants.