A Numerical Approach to Understanding Oscillator Neural Networks

Networks of coupled oscillators are a form of dynamical network originally inspired by various biological and physical systems. Much existing research focuses on simple, restricted models which are analytically tractable. In order to explore networks which do not fit into the existing models, we have programmed a flexible numerical simulation of oscillator networks. We have used the simulated network as an artificial neural network which uses a genetic algorithm for training. Preliminary results with the current neural network simulation suggest that the genetic algorithm is able to produce and optimize networks which perform differentiation between input frequencies. In future work, we hope to expand the existing numerical model, explore the role of noise during training on robustness, look at the effect of different network architectures, and explore optimization of neural networks for other tasks.