Review for NeurIPS paper: Efficient Variational Inference for Sparse Deep Learning with Theoretical Guarantee

They serve as proper proof of concept that the approach works in principle, but do the authors have an intuition as to how well their approach scales to larger models?