Reinforcement learning for neural networks using swarm intelligence

In this paper, we propose a swarm intelligence based reinforcement learning (SWIRL) method to train artificial neural networks (ANN). Basically, two swarm intelligence based algorithms are combined together to train the ANN models. Ant Colony Optimization (ACO) is applied to select ANN topology, while Particle Swarm Optimization (PSO) is applied to adjust ANN connection weights. To evaluate the performance of the SWIRL model, it is applied to double pole problem and robot localization through reinforcement learning. Extensive simulation results successfully demonstrate that SWIRL offers performance that is competitive with modern neuroevolutionary techniques, as well as its viability for real-world problems.