Evolving-Graph Gaussian Processes

David Blanco-Mulero; Markus Heinonen; Ville Kyrki

Evolving-Graph Gaussian Processes

David Blanco-Mulero^*, Markus Heinonen, Ville Kyrki

^*Corresponding author for this work

Research output: Contribution to conference › Paper › Scientific › peer-review

41 Downloads (Pure)

Abstract

Graph Gaussian Processes (GGPs) provide a dataefficient solution on graph structured domains. Existing approaches have focused on static structures, whereas many real graph data represent a dynamic structure, limiting the applications of GGPs. To overcome this we propose evolvingGraph Gaussian Processes (e-GGPs). The proposed method is capable of learning the transition function of graph vertices over time with a neighbourhood kernel to model the connectivity and interaction changes between vertices. We assess
the performance of our method on time-series regression problems where graphs evolve over time. We demonstrate the benefits of e-GGPs over static graph Gaussian Process approaches.

Original language	English
Number of pages	6
Publication status	Published - Jul 2021
MoE publication type	Not Eligible
Event	International Conference on Machine Learning: Time Series Workshop - Virtual, Online Duration: 24 Jul 2021 → 24 Jul 2021 http://roseyu.com/time-series-workshop/ https://roseyu.com/time-series-workshop/

Workshop

Workshop	International Conference on Machine Learning: Time Series Workshop
Abbreviated title	TSW-ICML
City	Virtual, Online
Period	24/07/2021 → 24/07/2021
Internet address	http://roseyu.com/time-series-workshop/ https://roseyu.com/time-series-workshop/

Keywords

gaussian process
Time series
Graph-based learning
probabilistic models

Access to Document

TSW-ICML2021_paper_21Accepted author manuscript, 2 MB

OpenUrl availability

Full text

Cite this

@conference{3f2dfaaddce849db9d8be1fb09189533,

title = "Evolving-Graph Gaussian Processes",

abstract = "Graph Gaussian Processes (GGPs) provide a dataefficient solution on graph structured domains. Existing approaches have focused on static structures, whereas many real graph data represent a dynamic structure, limiting the applications of GGPs. To overcome this we propose evolvingGraph Gaussian Processes (e-GGPs). The proposed method is capable of learning the transition function of graph vertices over time with a neighbourhood kernel to model the connectivity and interaction changes between vertices. We assessthe performance of our method on time-series regression problems where graphs evolve over time. We demonstrate the benefits of e-GGPs over static graph Gaussian Process approaches.",

keywords = "gaussian process, Time series, Graph-based learning, probabilistic models",

author = "David Blanco-Mulero and Markus Heinonen and Ville Kyrki",

year = "2021",

month = jul,

language = "English",

note = "International Conference on Machine Learning: Time Series Workshop, TSW-ICML ; Conference date: 24-07-2021 Through 24-07-2021",

url = "http://roseyu.com/time-series-workshop/, https://roseyu.com/time-series-workshop/",

}

TY - CONF

T1 - Evolving-Graph Gaussian Processes

AU - Blanco-Mulero, David

AU - Heinonen, Markus

AU - Kyrki, Ville

PY - 2021/7

Y1 - 2021/7

N2 - Graph Gaussian Processes (GGPs) provide a dataefficient solution on graph structured domains. Existing approaches have focused on static structures, whereas many real graph data represent a dynamic structure, limiting the applications of GGPs. To overcome this we propose evolvingGraph Gaussian Processes (e-GGPs). The proposed method is capable of learning the transition function of graph vertices over time with a neighbourhood kernel to model the connectivity and interaction changes between vertices. We assessthe performance of our method on time-series regression problems where graphs evolve over time. We demonstrate the benefits of e-GGPs over static graph Gaussian Process approaches.

AB - Graph Gaussian Processes (GGPs) provide a dataefficient solution on graph structured domains. Existing approaches have focused on static structures, whereas many real graph data represent a dynamic structure, limiting the applications of GGPs. To overcome this we propose evolvingGraph Gaussian Processes (e-GGPs). The proposed method is capable of learning the transition function of graph vertices over time with a neighbourhood kernel to model the connectivity and interaction changes between vertices. We assessthe performance of our method on time-series regression problems where graphs evolve over time. We demonstrate the benefits of e-GGPs over static graph Gaussian Process approaches.

KW - gaussian process

KW - Time series

KW - Graph-based learning

KW - probabilistic models

UR - https://github.com/dblanm/evolving-ggp

M3 - Paper

T2 - International Conference on Machine Learning: Time Series Workshop

Y2 - 24 July 2021 through 24 July 2021

ER -

Evolving-Graph Gaussian Processes

Abstract

Workshop

Keywords

Access to Document

OpenUrl availability

Other files and links

Fingerprint

-: AI spider silk threading

Cite this

Evolving-Graph Gaussian Processes

Abstract

Workshop

Keywords

Access to Document

OpenUrl availability

Other files and links

Fingerprint

Projects

-: AI spider silk threading

Cite this