We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Numerical Analysis

Title: Active Uncertainty Calibration in Bayesian ODE Solvers

Abstract: There is resurging interest, in statistics and machine learning, in solvers for ordinary differential equations (ODEs) that return probability measures instead of point estimates. Recently, Conrad et al. introduced a sampling-based class of methods that are 'well-calibrated' in a specific sense. But the computational cost of these methods is significantly above that of classic methods. On the other hand, Schober et al. pointed out a precise connection between classic Runge-Kutta ODE solvers and Gaussian filters, which gives only a rough probabilistic calibration, but at negligible cost overhead. By formulating the solution of ODEs as approximate inference in linear Gaussian SDEs, we investigate a range of probabilistic ODE solvers, that bridge the trade-off between computational cost and probabilistic calibration, and identify the inaccurate gradient measurement as the crucial source of uncertainty. We propose the novel filtering-based method Bayesian Quadrature filtering (BQF) which uses Bayesian quadrature to actively learn the imprecision in the gradient measurement by collecting multiple gradient evaluations.
Comments: 10 pages, 3 figures, accepted at UAI 2016
Subjects: Numerical Analysis (math.NA); Machine Learning (cs.LG); Machine Learning (stat.ML)
Cite as: arXiv:1605.03364 [cs.NA]
  (or arXiv:1605.03364v1 [cs.NA] for this version)

Submission history

From: Hans Kersting [view email]
[v1] Wed, 11 May 2016 10:24:04 GMT (385kb)
[v2] Mon, 23 May 2016 13:56:57 GMT (386kb)
[v3] Sat, 3 Nov 2018 09:39:02 GMT (143kb)

Link back to: arXiv, form interface, contact.