AI researcher Richard Ngo debates the claim that high-dimensional spaces rarely trap machine learning models in local minima · Digg