Skip to main content
The ARC Training Centre for
Transforming Maintenance through Data Science
Presentations

An introduction to time series analysis with reservoircomputing

braden-thorne

UWA (online)

Tuesday 8 February 2022

Reservoir computing is a framework for supervised machine learning tasks.

The underlying idea is to take some time series and inject it into a high dimensional recurrent network in order to construct a representation of the data in the reservoir state space.

Up to this point, research has focused on ways of training this state space to desired outputs or labels, and has done so with impressive results across many tasks and data sets.

However, the reservoir state space in and of itself is rich with information that can inform us about the original data stream without the need for training.

The ability to use reservoir computing in such an unsupervised way opens up new possibilities for the field, particularly for time series analysis.

Here, I'll introduce in more detail what reservoir time series analysis is and present a couple of ad-hoc methods for generating representative features from the reservoir state space.

I'll focus on the critical task of distinguishing signals from systems with varying parameters, presenting a testing framework for determining the capability of methods for this task.

This framework will allow us to highlight the strength and generality of the ad-hoc methods with respect to more well known time series analysis techniques, such as Fourier or recurrence quantification analysis.

Finally, we'll motivate the use of reservoir time series analysis in application with a short case study looking at distinguishing fault modes in bearings from vibration data.