New paper on SLANG: Fast Structured Covariance Approximations for Bayesian Deep Learning with Natural Gradient out. Will be presented at NeurIPS 2018.