Bayesian Learning via Neural Schrödinger-Föllmer Flows

Francisco Vargas; Andrius Ovsianas; David Fernandes; Mark Girolami; Neil D. Lawrence; Nikolas Nüsken

Back to publications

Bayesian Learning via Neural Schrödinger-Föllmer Flows

Francisco Vargas, Andrius Ovsianas, David Fernandes, Mark Girolami, Neil D. Lawrence, Nikolas Nüsken

Fourth Symposium on Advances in Approximate Bayesian Inference, 2022.

Abstract

In this work we explore a new framework for approximate Bayesian inference in large datasets based on stochastic control (i.e. Schrödinger bridges). We advocate stochastic control as a finite time and low variance alternative to popular steady-state methods such as stochastic gradient Langevin dynamics (SGLD). Furthermore, we discuss and adapt the existing theoretical guarantees of this framework and establish connections to already existing VI routines in SDE-based models.

Links

Cite this Paper

BibTeX


@InProceedings{publications/bayesian-learning-via-neural-schrödinger-föllmer-flows,
  title = 	 {Bayesian Learning via Neural Schrödinger-Föllmer Flows},
  author = 	 {Vargas, Francisco and Ovsianas, Andrius and Fernandes, David and Girolami, Mark and Lawrence, Neil D. and Nüsken, Nikolas},
  booktitle = 	 {Fourth Symposium on Advances in Approximate Bayesian Inference},
  year = 	 {2022},
  url = 	 {/publications/bayesian-learning-via-neural-schr%C3%B6dinger-f%C3%B6llmer-flows.html},
  abstract = 	 {In this work we explore a new framework for approximate Bayesian inference in large datasets based on stochastic control (i.e. Schrödinger bridges). We advocate stochastic control as a finite time and low variance alternative to popular steady-state methods such as stochastic gradient Langevin dynamics (SGLD). Furthermore, we discuss and adapt the existing theoretical guarantees of this framework and establish connections to already existing VI routines in SDE-based models.}
}

Endnote

%0 Conference Paper
%T Bayesian Learning via Neural Schrödinger-Föllmer Flows
%A Francisco Vargas
%A Andrius Ovsianas
%A David Fernandes
%A Mark Girolami
%A Neil D. Lawrence
%A Nikolas Nüsken
%B Fourth Symposium on Advances in Approximate Bayesian Inference
%D 2022	
%F publications/bayesian-learning-via-neural-schrödinger-föllmer-flows
%U /publications/bayesian-learning-via-neural-schr%C3%B6dinger-f%C3%B6llmer-flows.html
%X In this work we explore a new framework for approximate Bayesian inference in large datasets based on stochastic control (i.e. Schrödinger bridges). We advocate stochastic control as a finite time and low variance alternative to popular steady-state methods such as stochastic gradient Langevin dynamics (SGLD). Furthermore, we discuss and adapt the existing theoretical guarantees of this framework and establish connections to already existing VI routines in SDE-based models.

RIS


TY  - CPAPER
TI  - Bayesian Learning via Neural Schrödinger-Föllmer Flows
AU  - Francisco Vargas
AU  - Andrius Ovsianas
AU  - David Fernandes
AU  - Mark Girolami
AU  - Neil D. Lawrence
AU  - Nikolas Nüsken
BT  - Fourth Symposium on Advances in Approximate Bayesian Inference
DA  - 2022/02/01	
ID  - publications/bayesian-learning-via-neural-schrödinger-föllmer-flows
UR  - /publications/bayesian-learning-via-neural-schr%C3%B6dinger-f%C3%B6llmer-flows.html
AB  - In this work we explore a new framework for approximate Bayesian inference in large datasets based on stochastic control (i.e. Schrödinger bridges). We advocate stochastic control as a finite time and low variance alternative to popular steady-state methods such as stochastic gradient Langevin dynamics (SGLD). Furthermore, we discuss and adapt the existing theoretical guarantees of this framework and establish connections to already existing VI routines in SDE-based models.
ER  -

APA


Vargas, F., Ovsianas, A., Fernandes, D., Girolami, M., Lawrence, N.D. & Nüsken, N.. (2022). Bayesian Learning via Neural Schrödinger-Föllmer Flows. Fourth Symposium on Advances in Approximate Bayesian Inference Available from /publications/bayesian-learning-via-neural-schr%C3%B6dinger-f%C3%B6llmer-flows.html.