I’m going through a tensorflow tutorial in which they make a simple model. they’ve gotten to the point where they are explaining the importance of optimizing your W and b variables with respect to gradient decent.
I thought, “Oh, that’s a heuristic, and it’s blind. HTM modifies itself in an intelligent way, where it can see exactly why it modified itself (the particular patterns that were recognized in the particular context it’s in).”
So, is it safe to say that HTM excels because it’s ‘optimization’ method is ‘better informed’ than the optimization method of neural nets (gradient descent)? Or is my intuition a little naive?