Aol PhD Fellows, Andreas Veit and Michael Wilber, traveled to Barcelona, Spain to present their paper at the Conference on Neural Information Processing Systems (NIPS 2016). Their paper, “Residual Networks Behave Like Ensembles of Relatively Shallow Networks,” investigates residual networks, state-of-the-art models used for computer vision and machine learning. They discover that residual networks are easy to train because they can be expressed as a collection of many short paths. Residual networks avoid the vanishing gradient problem by introducing short paths which can carry gradient throughout the extent of very deep networks.
Cornell Tech 2 West Loop Road NY NY 10044