Revealing Functions of Extra-Large Excitatory Postsynaptic Potentials: Insights from Dynamical Characteristics of Reservoir Computing with Spiking Neural Networks

Asato Fujimoto, Sou Nobukawa, Yusuke Sakemi, Yoshiho Ikeuchi, Kazuyuki Aihara

Abstract: Extensive neuroscience studies have been conducted to unravel mechanisms behind the emergence of neural activity in physiological neural systems. Long-tailed excitatory postsynaptic potentials (EPSPs), involving a minority of extra-large (XL) EPSPs, are currently garnering much attention, which strongly relates to cognitive functions. In addition to physiological studies, mathematical modeling approaches are effective in neuroscience because they can incorporate experimentally revealed characteristics into mathematical models to reveal the structural factors to produce the neural dynamics. Reservoir computing (RC) with spiking neural networks (SNNs) has recently been explored as an approach to elucidate how brain structures generate functions by investigating the functionality of the network features. In this context, the purpose of this study is to reveal the role of XL EPSP at the functional level by analyzing XL-EPSP-induced dynamics. This study evaluated the spatio-temporal patterns of neural activity and the learning performance of RC with SNNs incorporating XL EPSPs. Results showed that RC performance exhibited an inverted U-shape according to the mutual interaction strength between neural populations. In summary, an optimal degree of XL EPSP induces appropriate mutual interneural interactions that reflect the input stimulus; consequently, this effect contributes to learning performance in RC. RC modeling offers a possible solution for uncovering the functional aspects of structural characteristics within physiological neural networks.