Data-Driven Experimental Design Team Online Seminar

This is an online seminar by the Data-Driven Experimental Design Team at RIKEN's Center for Advanced Intelligence Project (AIP). In this session, Shuhei Sugiura presents a talk titled "On the universality of uniform approximation in reservoir computing." Reservoir computing (RC) is a method for approximating dynamical systems. It learns at low computational cost by training only the static readout function over the model's internal state. The reservoir, which needs no training, can be implemented in hardware using various physical phenomena, offering improved energy efficiency. The talk proposes necessary and sufficient conditions for uniform approximation by RC to hold. Known sufficient conditions for universality require the reservoir to have a single stable state, which excludes many physical implementations. This research presents conditions that also apply to reservoirs without a single stable state, and confirms their importance in realistic settings.