Summary
International Symposium on Nonlinear Theory and Its Applications
2022
Session Number:B3L-D
Session:
Number:B3L-D-05
Investigation of Metrics for Readout-Independent Evaluation of the Functional Robustness of Liquid State Machines
Naoki Hirata , Naoki Wakamiya,
pp.347-350
Publication Date:12/12/2022
Online ISSN:2188-5079
Summary:
Physical reservoir computing, which is the physical implementation of reservoir computing models, has attracted attention of researchers. Although it has a variety of applications, as a physical entity, it is prone to physical failures. Therefore, it is necessary to design a reservoir which is functionally robust to failures. More specifically, it is desired that a reservoir maintains the dynamics even after failures. For this purpose, in this paper, we investigate metrics to evaluate the functional robustness of a reservoir. By comparing distance metrics, we found that D-spike and D-interval were suitable for such evaluation.