Rao-Blackwellised Reparameterisation Gradients

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Lam, Kevin H., Bui, Thang D., Deligiannidis, George, Teh, Yee Whye
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917025569308672
author Lam, Kevin H.
Bui, Thang D.
Deligiannidis, George
Teh, Yee Whye
author_facet Lam, Kevin H.
Bui, Thang D.
Deligiannidis, George
Teh, Yee Whye
contents Latent Gaussian variables have been popularised in probabilistic machine learning. In turn, gradient estimators are the machinery that facilitates gradient-based optimisation for models with latent Gaussian variables. The reparameterisation trick is often used as the default estimator as it is simple to implement and yields low-variance gradients for variational inference. In this work, we propose the R2-G2 estimator as the Rao-Blackwellisation of the reparameterisation gradient estimator. Interestingly, we show that the local reparameterisation gradient estimator for Bayesian MLPs is an instance of the R2-G2 estimator and Rao-Blackwellisation. This lets us extend benefits of Rao-Blackwellised gradients to a suite of probabilistic models. We show that initial training with R2-G2 consistently yields better performance in models with multiple applications of the reparameterisation trick.
format Preprint
id arxiv_https___arxiv_org_abs_2506_07687
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Rao-Blackwellised Reparameterisation Gradients
Lam, Kevin H.
Bui, Thang D.
Deligiannidis, George
Teh, Yee Whye
Machine Learning
Latent Gaussian variables have been popularised in probabilistic machine learning. In turn, gradient estimators are the machinery that facilitates gradient-based optimisation for models with latent Gaussian variables. The reparameterisation trick is often used as the default estimator as it is simple to implement and yields low-variance gradients for variational inference. In this work, we propose the R2-G2 estimator as the Rao-Blackwellisation of the reparameterisation gradient estimator. Interestingly, we show that the local reparameterisation gradient estimator for Bayesian MLPs is an instance of the R2-G2 estimator and Rao-Blackwellisation. This lets us extend benefits of Rao-Blackwellised gradients to a suite of probabilistic models. We show that initial training with R2-G2 consistently yields better performance in models with multiple applications of the reparameterisation trick.
title Rao-Blackwellised Reparameterisation Gradients
topic Machine Learning
url https://arxiv.org/abs/2506.07687