Reviews: Neural Architecture Search with Bayesian Optimisation and Optimal Transport

The paper describes a new distance metric for neural networks based on optimal transport. Based on this distance, the authors introduce a new Bayesian optimization strategy to optimize the architecture of neural networks. Overall I think the methods is interesting and that the proposed distance might be also of interest for other applications than Bayesian optimization. However, it seems that computing the distance also requires some manual tuning such as defining the cost matrix which in turn requires expert knowledge. My only point of criticism is that this might hinder its success in automated machine learning applications, where human interaction should be reduced to a minimum.