Traversing Time with Multi-Resolution Gaussian Process State-Space Models

Krista Elena Longi, Jakob Lindinger, Olaf Duennbier, Melih Kandemir, Arto Klami, Barbara Rakitsch

Forskningsoutput: Kapitel i bok/rapport/konferenshandlingKonferensbidragVetenskapligPeer review

Sammanfattning

Gaussian Process state-space models capture complex temporal dependencies in a principled manner by placing a Gaussian Process prior on the transition function. These models have a natural interpretation as discretized stochastic differential equations, but inference for long sequences with fast and slow transitions is difficult. Fast transitions need tight discretizations whereas slow transitions require backpropagating the gradients over long subtrajectories. We propose a novel Gaussian process state-space architecture composed of multiple components, each trained on a different resolution, to model effects on different timescales. The combined model allows traversing time on adaptive scales, providing efficient inference for arbitrarily long sequences with complex dynamics. We benchmark our novel method on semi-synthetic data and on an engine modeling task. In both experiments, our approach compares favorably against its state-of-the-art alternatives that operate on a single time-scale only.
Originalspråkengelska
Titel på värdpublikationProceedings of The 4th Annual Learning for Dynamics and Control Conference, PMLR
RedaktörerRoya Firoozi, Negar Mehr, Esen Yel, Rika Antonova, Jeannette Bohg, Mac Schwager, Mykel Kochenderfer
FörlagJournal of Machine Learning Research
Utgivningsdatum2022
Sidor366-377
DOI
StatusPublicerad - 2022
MoE-publikationstypA4 Artikel i en konferenspublikation
EvenemangAnnual Learning for Dynamics and Control Conference - Stanford, CA, Förenta Staterna (USA)
Varaktighet: 23 juni 202224 juni 2022
Konferensnummer: 4

Publikationsserier

NamnProceedings of Machine Learning Research (PMLR)
Volym168
ISSN (elektroniskt)2640-3498

Vetenskapsgrenar

  • 113 Data- och informationsvetenskap

Citera det här