Where would you say the controversy comes from? More or less controversial then other purely theoretical neuroscience that don't (afaik) have supporting empirical evidence?

Thing is that the people arguing pro GD (e.g. me) are far more convinced about it than those of other theories. The correlation argument I am making requires just local linearity and under that assumption becomes provable. Hence, to us, it is not really a theory.

This is completely misrepresenting the thread @KordingLab . You made slippery arguments which people questioned, then you stated that you weren't arguing for GD or a mechanism..and now you are presenting it as people were questioning GD.

But I never in the thread did argue for GD as a mechanism!

Ahem.... see Equation 1: pnas.org/content/116/21…

yes, it is awesome ;)

@dileeplearning does remind me that the controversy may have been rather due to misunderstandings. GD may, indeed, be trivially correlated to updates in the brain.

I’m puzzled as, surveying the landscape visible from Dunning & Kruger’s Mount Stupid, I’m not aware of an AI program that works without some kind of gradient descent (as pure solution search takes too long) which suggests biological computation will have the same limit.

But couldn't biologists argue that the reason could be that simply all AI systems share a common evolutionary basis (code written by humans) and may thus be fundamentally different.

Frankly, I am a bit puzzled by most counter-arguments. Our best physical theories are mathematical *descriptions* of reality. If you take Newton’s inverse-square law for example this describes experimental phenomena but no astronomer thinks that planets do calculations.

There are several lines of reasoning to justify the gradient formalism. First, the argument of practicality. We now have very good automatic differentiation software that can revolutionise the theoretical sciences. Why don’t theoretical neuroscientists make good use of this?