Poster Session

The Analysis of Time-Delay Reservoir Computing Focusing on Dynamical Properties of Delay Differential Equations.

Ikuhide Kinoshita
The University of Tokyo
Co-Author(s):    Akihiko Akao, Sho Shirasaka, Kiyoshi Kotani, Yasuhiko Jimbo
Reservoir Computing (RC) is a machine-learning paradigm that is capable to process empirical time-series data. This paradigm is based on a neural network with a fixed hidden layer owning a high-dimensional state space, called a reservoir. Reservoirs including time-delays are considered to be good candidates for practical applications because they make hardware realization of the high-dimensional reservoirs simple. Performance of the well-trained RCs depends both on dynamical properties of attractors of the reservoirs and tasks they solve. Therefore, in the conventional RCs, there arise task-wise optimization problems of the reservoirs, which have been solved based on trial and error approaches. This means that it is necessary to switch reservoirs for the optimization of the performance for each task. In this study, we compared and investigated the time-delay RCs method for various tasks, focusing on the dynamical properties of reservoirs and using multiple time-delay nodes. Specifically, the performance in various tasks was measured using reservoirs having different dynamic characteristics arranged in parallel. These findings pave the way to a more versatile and robust time-delay RCs.