Network of Theseus (like the ship)

Vighnesh Subramaniam, Colin Conwell, Boris Katz +2

A standard assumption in deep learning is that the inductive bias introduced by a neural network architecture must persist from training through inference. The architecture you tra…