Imperial College London
Portrait Picture

ProfessorThomasCass

Faculty of Natural Sciences, Department of Mathematics

Professor of Mathematics
 
 
 
//

Contact

 

thomas.cass

 
 
//

Location

 

808Weeks BuildingSouth Kensington Campus

//

Summary

 

Publications

Citation

BibTex format

@inproceedings{Lemercier:2021,
author = {Lemercier, M and Salvi, C and Cass, T and Bonilla, EV and Damoulas, T and Lyons, T},
pages = {6233--6242},
publisher = {PMLR},
title = {SigGPDE: scaling sparse Gaussian processes on sequential data},
url = {http://arxiv.org/abs/2105.04211v1},
year = {2021}
}
Download

RIS format (EndNote, RefMan)

TY  - CPAPER
AB  - Making predictions and quantifying their uncertainty when the input data is sequential is a fundamental learning challenge, recently attracting increasing attention. We develop SigGPDE, a new    scalable sparse variational inference framework for Gaussian Processes (GPs) on sequential data. Our contribution is twofold.  First, we construct inducing variables underpinning the sparse approximation so that the resulting evidence lower bound (ELBO) does not require any matrix inversion.  Second, we show that the gradients of the GP signature kernel are solutions of a hyperbolic partial differential equation (PDE). This theoretical insight allows us to build an efficient back-propagation algorithm to optimize the ELBO. We showcase the significant computational gains of SigGPDE compared to existing methods, while achieving state-of-the-art performance for classification tasks on large datasets of up to1millionmultivariate time series.
AU  - Lemercier,M
AU  - Salvi,C
AU  - Cass,T
AU  - Bonilla,EV
AU  - Damoulas,T
AU  - Lyons,T
EP  - 6242
PB  - PMLR
PY  - 2021///
SN  - 2640-3498
SP  - 6233
TI  - SigGPDE: scaling sparse Gaussian processes on sequential data
UR  - http://arxiv.org/abs/2105.04211v1
UR  - https://proceedings.mlr.press/v139/lemercier21a.html
UR  - http://hdl.handle.net/10044/1/89749
ER  -
Download