I'm Not Enough in the Weeds

A lot of ma research still quite empirical. Justifying the correct learning rate optimizers still very difficult, or various tipes up optimizes and things like that. And so for me, it it still boils down to empirically what you can or can't verify. Or more so, ratherf you can choose something to work, can dive deeper into understanding why or why not. Perhaps that's a little more controversial.