Deep Learning - What Are Some of the Most Important Open Questions?

Deep learning is such a huge field these days that i'd be ignorant speaking about anybody else's. Can we find sparse networks retroactively at the beginning of training, and can we find them efficiently? And if not early in training, can we develop pruning techniques that work early in training? Initialization is ambitious. Let's start with you kno, a quarter of the way through training, and work our way there. I've got some work in progress on that subject that i'm interested in sharing as well where i've tried to bring a kind of a scientific edge to that literature. It's still early days this job market, but expect more an that, hopefully later in the