Konferenzbeitrag
Priors for Linear Differential Equations
Lade...
Volltext URI
Dokumententyp
Text/Conference Paper
Dateien
Zusatzinformation
Datum
2019
Autor:innen
Zeitschriftentitel
ISSN der Zeitschrift
Bandtitel
Verlag
Gesellschaft für Informatik e.V.
Zusammenfassung
We algorithmically construct multi-output Gaussian process priors which satisfy linear differential equations. We parametrize all solutions of the differential equations using Gröbner bases for controllable systems. If successful, a push forward along the parametrization is the desired prior. This prior yields an interpretable machine learning model, which can combine linear differential equations with noisy data points.