Statistics for Differential Topological Properties between Data Sets with an Application to Reservoir Computers

Abstract

It is common for researchers to record long, multiple time series from experiments or calculations. But sometimes there are no good models for the systems or no applicable mathematical theorems that can tell us when there are basic relationships between subsets of the time series data such as continuity, differentiability, embeddings, etc. The data is often higher dimensional and simple plotting will not guide us. At that point fitting the data to polynomials, Fourier series, etc. becomes uncertain. Even at the simplest level, having data that shows there is a function between the data subsets is useful and a negative answer means that more particular data fitting or analysis will be suspect and probably fail. We show here statistics that test time series subsets for basic mathematical properties and relations between them that not only indicate when more specific analyses are safe to do, but whether the systems are operating correctly. We apply these statistics to examples from reservoir computing where an important property of reservoir computers is that the reservoir system establishes an embedding of the drive system in order to make any other calculations with the reservoir computer successful.