Bayesian learning via neural Schrödinger–Föllmer flows

Francisco Vargas; Andrius Ovsianas; David Fernandes; Mark Girolami; Neil Lawrence; Nikolas Nusken

Bayesian learning via neural Schrödinger–Föllmer flows

Francisco Vargas, Andrius Ovsianas, David Fernandes, Mark Girolami, Neil Lawrence, Nikolas Nusken

Research output: Contribution to journal › Article › peer-review

60 Downloads (Pure)

Abstract

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

Original language	English
Journal	STATISTICS AND COMPUTING
Volume	33
Publication status	Published - 23 Nov 2022

Access to Document

Bayesian learning via neural Schroedinger Foellmer flowsAccepted author manuscript, 4.87 MB

Cite this

@article{dc44c48b0ace4623ab1e780a9ea0959b,

title = "Bayesian learning via neural Schr{\"o}dinger–F{\"o}llmer flows",

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

author = "Francisco Vargas and Andrius Ovsianas and David Fernandes and Mark Girolami and Neil Lawrence and Nikolas Nusken",

year = "2022",

month = nov,

day = "23",

language = "English",

volume = "33",

journal = "STATISTICS AND COMPUTING",

issn = "0960-3174",

publisher = "Springer Netherlands",

}

TY - JOUR

T1 - Bayesian learning via neural Schrödinger–Föllmer flows

AU - Vargas, Francisco

AU - Ovsianas, Andrius

AU - Fernandes, David

AU - Girolami, Mark

AU - Lawrence, Neil

AU - Nusken, Nikolas

PY - 2022/11/23

Y1 - 2022/11/23

N2 - In this work we explore a new framework for approximate Bayesian inference in large datasets based on stochastic control. We advocate stochastic control as a finite time and low variance alternative to popular steady-state methods such as stochastic gradient Langevin dynamics. Furthermore, we discuss and adapt the existing theoretical guarantees of this framework and establish connections to already existing VI routines in SDE-based models.

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

M3 - Article

SN - 0960-3174

VL - 33

JO - STATISTICS AND COMPUTING

JF - STATISTICS AND COMPUTING

ER -

Bayesian learning via neural Schrödinger–Föllmer flows

Abstract

Access to Document

Fingerprint

Cite this