In this work we propose a novel interpretation of residual networks showing that they can be seen as a collection of many paths of differing length. Moreover, residual networks seem to enable very deep networks by leveraging only the short paths during training. To support this observation, we rewrite residual networks as an explicit collection of paths. Unlike traditional models, paths through residual networks vary in length. Further, a lesion study reveals that these paths show ensemble-like behavior in the sense that they do not strongly depend on each other. Finally, and most surprising, most paths are shorter than one might expect, and only the short paths are needed during training, as longer paths do not contribute any gradient. For example, most of the gradient in a residual network with 110 layers comes from paths that are only 10-34 layers deep. Our results reveal one of the key characteristics that seem to enable the training of very deep networks: Residual networks avoid the vanishing gradient problem by introducing short paths which can carry gradient throughout the extent of very deep networks.

I think the experiments in this paper are very interesting and reveal interesting properties of Resnet. However, I don't think the claim that ResNets are "exponential ensembles" has been made adequetely. If the claim were weakened to "behave like exponential ensembles", this would seem to be reasonably well supported by the experiments, and it's also an interesting result. But I think the question of why this happens is still open. One thing that stands out is that no definition of "ensemble" is ever given, not even a rough definition.