In today's ML world, theory follows practice. You have to demonstrate SOTA results in a problem area before people will pay attention to your theory.

So, if you're proposing some theory, first implement it and demonstrate it. Otherwise, it's just vaporware.