Online Gaussian Process regression with non-Gaussian likelihood
| Autor: | David Seiferth, Maximilian Mühlegg, Girish Chowdhary, Florian Holzapfel |
|---|---|
| Rok vydání: | 2017 |
| Předmět: |
Pointwise
Mathematical optimization Gaussian 010501 environmental sciences 01 natural sciences Statistics::Computation symbols.namesake Bayes' theorem ComputingMethodologies_PATTERNRECOGNITION Kriging Laplace's method Expectation propagation 0103 physical sciences symbols Gaussian function Statistics::Methodology 010306 general physics Gaussian process Algorithm 0105 earth and related environmental sciences Mathematics |
| Zdroj: | ACC |
| DOI: | 10.23919/acc.2017.7963429 |
| Popis: | We present a new algorithm for GP regression over data with non-Gaussian likelihood that does not require costly MCMC sampling, or variational Bayes optimization. In our method, which we term Meta-GP, we model the likelihood by another Gaussian Process point-wise in time. This approach allows for the calculation of the posterior predictive mean and variance in an analytical way pointwise in time, leading to an online inference algorithm. As a result, our method can work with streaming data, is analytically tractable, computationally efficient while being as accurate or better than Expectation Propagation, Laplace Approximation, and MCMC inference methods for non-Gaussian likelihood data. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|