Fundamental problems elude many strains of deep learning, says LeCun, including the mystery of how to measure information.
