Toggle Main Menu Toggle Search

Open Access padlockePrints

Composable Models for Online Bayesian Analysis of Streaming Data

Lookup NU author(s): Jonathan Law, Professor Darren Wilkinson

Downloads


Licence

This work is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).


Abstract

Data is rapidly increasing in volume and velocityand the Internet of Things (IoT) is one importantsource of this data. The IoT is a collection of connecteddevices (things) which are constantly recordingdata from their surroundings using on-board sensors.These devices can record and stream data to thecloud at a very high rate, leading to high storage andanalysis costs. In order to ameliorate these costs, thedata is modelled as a stream and analysed online tolearn about the underlying process, perform interpolationand smoothing and make forecasts and predictions.Conventional state space modelling tools assumethe observations occur on a fixed regular time grid.However, many sensors change their sampling frequency,sometimes adaptively, or get interrupted and re-startedout of sync with the previous sampling grid, or justgenerate event data at irregular times. It is thereforedesirable to model the system as a partially and irregularlyobserved Markov process which evolves incontinuous time. Both the process and the observationmodel are potentially non-linear. Particle filters thereforerepresent the simplest approach to online analysis.A functional Scala library of composable continuoustime Markov process models has been developed in orderto model the wide variety of data captured in theIoT.


Publication metadata

Author(s): Law J, Wilkinson DJ

Publication type: Article

Publication status: Published

Journal: Statistics and Computing

Year: 2018

Volume: 28

Issue: 6

Pages: 1119-1137

Print publication date: 01/11/2018

Online publication date: 31/10/2017

Acceptance date: 22/09/2017

Date deposited: 22/09/2017

ISSN (print): 0960-3174

ISSN (electronic): 1573-1375

Publisher: Springer

URL: https://doi.org/10.1007/s11222-017-9783-1

DOI: 10.1007/s11222-017-9783-1


Altmetrics

Altmetrics provided by Altmetric


Actions

Find at Newcastle University icon    Link to this publication


Share