Fetching the paper…

Evaluating representations by the complexity of learning low-loss predictors · Around