Rao-Blackwellised reparameterisation gradients

Conference item

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.

Authors: Lam, K; Bui, T; Deligiannidis, G; Teh, YW

• Publication status: Accepted
• Peer review status: Peer reviewed
• Publisher: Neural Information Processing Systems Foundation
• Publication date: 2025-06-09
• Acceptance date: 2025-09-18
• Event title: Thirty-Ninth Annual Conference on Neural Information Processing Systems
• Event location: San Diego, California, USA and Mexico City, Mexico
• Event website: https://neurips.cc/
• Event start date: 2025-11-30
• Event end date: 2025-12-07
• Language: English