Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Generalized Variational Inference in Function Spaces: Gaussian Measures meet Bayesian Deep Learning

About

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 optimisation criterion and 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 parametrisation 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.

Veit D. Wild, Robert Hu, Dino Sejdinovic• 2022

Related benchmarks

TaskDatasetResultRank
RegressionUCI NAVAL (test)
Negative Log Likelihood-7.21
42
RegressionBoston UCI (test)--
36
RegressionUCI Red Wine 10% (test)
NLL0.26
9
RegressionUCI Energy 10% (test)
NLL0.54
9
RegressionUCI Yacht (10% test)
NLL0.1
9
RegressionUCI Concrete 10% (test)
NLL2.64
9
RegressionUCI Kin8nm 10% (test)
NLL-1.2
8
RegressionUCI Power 10% (test)
NLL2.69
8
RegressionUCI Protein 10% (test)
NLL2.85
8
Showing 9 of 9 rows

Other info

Follow for update