Projektbeschreibung
Nichtparametrischer Bayes'scher Ansatz für die Variablenauswahl bei hochdimensionalen kausalen Inferenzen
Die kausale Inferenz kann sich bei größeren Datenmengen schwieriger gestalten. In hochdimensionalen Umgebungen ist die Einbeziehung aller Variablen unmöglich, während die Einbeziehung von zu wenigen Variablen möglicherweise zu falschen Ergebnissen führt. Daher ist die Auswahl der Variablen unbedingt erforderlich, doch die verfügbaren Methoden sind begrenzt. Das vom ERC finanzierte Projekt BayCause zielt darauf ab, nichtparametrische Bayes'sche Methoden und Theorien zu entwickeln, die für die Variablenauswahl bei hochdimensionalen kausalen Inferenzen geeignet sind. Jüngste theoretische Fortschritte in der Bernstein-von-Mises-Theorie und der hochdimensionalen nichtparametrischen Regression ermöglichten die Anwendung von nichtparametrischen Bayes'schen Ansätzen bei der kausalen Inferenz. Die im Rahmen des Projekts entwickelten Methoden werden den begrenzten Rahmen, in dem zuverlässige hochdimensionale Kausalschlüsse möglich sind, erweitern und zu Anwendungen in Medizin, Wirtschaft und anderen Bereichen führen.
Ziel
Causal conclusions are at the center of research, yet notoriously difficult to obtain. Many research studies report correlations only, which, in line with the maxim, do not imply causation. With correlations, one can make predictions. With causation, one can intervene.
Paradoxically, causal inference can become harder when more data becomes available. In the by now increasingly common high-dimensional settings which are the focus of this proposal, including all variables is impossible while including too few can severely bias results. Variable selection becomes necessary, yet available methods are in short supply.
My aim is to develop Bayesian nonparametric methods and theory for high-dimensional causal inference. Bayesian nonparametrics is eminently suited for variable selection in causal inference, because it excels at both incorporating and describing uncertainty. Recent theoretical advances, in particular in Bernstein-von Mises theory and high-dimensional nonparametric regression, have now finally opened up causal inference to Bayesian nonparametric approaches.
I will investigate high-dimensional versions of the two most important causal frameworks, based on unconfoundedness and directed acyclic graphs. I will focus on novel aspects scarcely available in the literature, including uncertainty quantification, a broad range of data types, and nonlinear relationships.
My expertise in causal inference, Bayesian nonparametrics, variable selection and survival analysis puts me in a unique position to work on this multifaceted challenge. My dual track in theoretical and applied statistics enables me to identify the problems which have highest priority in practice and are mathematically interesting. The novel methods with solid mathematical statistical foundation resulting from this proposal will tremendously expand the now limited settings in which trustworthy high-dimensional causal inference is possible, with applications in medicine, economics and many other fields.
Schlüsselbegriffe
Programm/Programme
- HORIZON.1.1 - European Research Council (ERC) Main Programme
Thema/Themen
Aufforderung zur Vorschlagseinreichung
(öffnet in neuem Fenster) ERC-2022-STG
Andere Projekte für diesen Aufruf anzeigenFinanzierungsplan
HORIZON-ERC -Gastgebende Einrichtung
1081 HV Amsterdam
Niederlande