Improving Anytime Prediction with Parallel Cascaded Networks and a Temporal-Difference Loss

Open in new window