[curiousgal]

I mean, is it really Monte Carlo? He's iteratively trying to optimize a function which makes this more like SGD, no?

[jasonheh]

It definitely felt that way to me.

When he was finding the optimal angle for a single sampled shot, that's the partial derivative of his probability of making the basket for that shot wrt the surface normal at the backboard hit position (or something like that).

If you average all of those you get a gradient descent update.