> but a more theoretical foundation, one that enables me to think and envisage new architectures, new techniques or combinations Advances in DL have been mostl

> but a more theoretical foundation, one that enables me to think and envisage new architectures, new techniques or combinations Advances in DL have been mostly empirical not theoretical from what I have seen. It's less about being able to "envisage" and more about empirical testing.