Ecole d'ingénieur et centre de recherche en Sciences du numérique

Exact Gaussian process regression with distributed computations

Nguyen, Duc-Trung; Filippone, Maurizio; Michiardi, Pietro

SAC 2019, 34th ACM/SIGAPP Symposium on Applied Computing, April 8-12, 2019, Limassol, Cyprus

Gaussian Processes (GPs) are powerful non-parametric Bayesian models for function estimation, but suffer from high complexity in terms of both computation and storage. To address such issues, approximation methods have flourished in the literature, including model approximations and approximate inference. However, these methods often sacrifice accuracy for scalability. In this work, we present the design and evaluation of a distributed method for exact GP inference, that achieves true model parallelism using simple, high-level distributed computing frameworks. Our experiments show that exact inference at scale is not only feasible, but it also brings substantial benefits in terms of low error rates and accurate quantification of uncertainty.

Document Doi Bibtex

Titre:Exact Gaussian process regression with distributed computations
Mots Clés:Regression, Matrix Factorization, Distributed computing
Type:Conférence
Langue:English
Ville:Limassol
Pays:CHYPRE
Date:
Département:Data Science
Eurecom ref:5851
Copyright: © ACM, 2019. This is the author's version of the work. It is posted here by permission of ACM for your personal use. Not for redistribution. The definitive version was published in SAC 2019, 34th ACM/SIGAPP Symposium on Applied Computing, April 8-12, 2019, Limassol, Cyprus http://dx.doi.org/10.1145/3297280.3297409
Bibtex: @inproceedings{EURECOM+5851, doi = {http://dx.doi.org/10.1145/3297280.3297409}, year = {2019}, title = {{E}xact {G}aussian process regression with distributed computations}, author = {{N}guyen, {D}uc-{T}rung and {F}ilippone, {M}aurizio and {M}ichiardi, {P}ietro}, booktitle = {{SAC} 2019, 34th {ACM}/{SIGAPP} {S}ymposium on {A}pplied {C}omputing, {A}pril 8-12, 2019, {L}imassol, {C}yprus }, address = {{L}imassol, {CHYPRE}}, month = {04}, url = {http://www.eurecom.fr/publication/5851} }
Voir aussi: