We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Statistics > Machine Learning

Title: Gradient Regularisation as Approximate Variational Inference

Abstract: Variational inference in Bayesian neural networks is usually performed using stochastic sampling which gives very high-variance gradients, and hence slow learning. Here, we show that it is possible to obtain a deterministic approximation of the ELBO for a Bayesian neural network by doing a Taylor-series expansion around the mean of the current variational distribution. The resulting approximate ELBO is the training-log-likelihood plus a squared gradient regulariser. In addition to learning the approximate posterior variance, we also consider a uniform-variance approximate posterior, inspired by the stationary distribution of SGD. The corresponding approximate ELBO has a simple form, as the log-likelihood plus a simple squared-gradient regulariser. We argue that this squared-gradient regularisation may at the root of the excellent empirical performance of SGD.
Subjects: Machine Learning (stat.ML); Machine Learning (cs.LG)
Cite as: arXiv:2011.10443 [stat.ML]
  (or arXiv:2011.10443v1 [stat.ML] for this version)

Submission history

From: Ali Unlu [view email]
[v1] Fri, 20 Nov 2020 15:16:18 GMT (111kb,D)
[v2] Tue, 10 Aug 2021 13:20:59 GMT (1584kb,D)

Link back to: arXiv, form interface, contact.