Abstract: Historically, particle swarm optimisation algorithms have successfully been applied to neural network training, often outperforming traditional gradient-based approaches. Studies have however shown that particle swarm optimisation algorithms do not scale very well, performing poorly on high dimensional neural network architectures. This study aims to research the effect of using a chaotic particle swarm optimisation algorithm for neural network training. The resulting neural network’s saturation and overfitting, as well as high-dimensional performance, will be investigated.