Delay Learning in a Biologically Plausible Reservoir Computing Model

Abstract

In this thesis, a reservoir model created by Paugam-Moisy et al. (2008) was replicated, and subsequently modified to be more biologically plausible. The use of a di erent neuron model and STDP function had no negative impact on performance. For the delay learning mechanism, replacing delay adaptation with delay selection proved viable, but did result in somewhat lower performance. It was discovered here that the reservoir in this model served no purpose. Therefore, the use of a small-world Watts-Strogatz reservoir instead of a randomly connected reservoir, could not affect performance.