Random deep neural networks are biased towards simple functions December 12, 2019conference preceedings with poster at NeurIPS 2019, Vancouver, CA