NeurIPS 2020

Efficient Low Rank Gaussian Variational Inference for Neural Networks


Meta Review

The authors explore the effects and quality of a variational approximation to the posterior of NNs based on low-rank Gaussian distributions per layer. Strengths: - the paper is well written and covers the literature well - the novel local reparameterization trick and low-rank variational approximation are sound and efficient Weaknesses: - missing comparisons to baselines (these were promised to be delivered upon acceptance)