Why Deep Learning Works at All: Yoshua Bengio on Composition, Saddle Points and the Case for Unsupervised Learning

This is the video's original transcript, shown as is and unedited. The edited full text, the summary and the translations are on the article page.