Timezone: »

 
Poster
Generalized Variational Inference in Function Spaces: Gaussian Measures meet Bayesian Deep Learning
Veit David Wild · Robert Hu · Dino Sejdinovic

Tue Nov 29 09:00 AM -- 11:00 AM (PST) @ Hall J #509
in Poster Session 1 »

We develop a framework for generalized variational inference in infinite-dimensional function spaces and use it to construct a method termed Gaussian Wasserstein inference (GWI). GWI leverages the Wasserstein distance between Gaussian measures on the Hilbert space of square-integrable functions in order to determine a variational posterior using a tractable optimization criterion. It avoids pathologies arising in standard variational function space inference. An exciting application of GWI is the ability to use deep neural networks in the variational parametrization of GWI, combining their superior predictive performance with the principled uncertainty quantification analogous to that of Gaussian processes. The proposed method obtains state-of-the-art performance on several benchmark datasets.

Keywords: Variational Inference · Gaussian Processes · bayesian inference · Bayesian Learning · Function Space Inference · Generalised Variational Inference · Gaussian Measures
[ Paper]  [ Poster]  [ OpenReview] 

Author Information

Veit David Wild (University of Oxford)
Robert Hu (Amazon)
Dino Sejdinovic (University of Adelaide)

More from the Same Authors