newtons method is gradient descent

@Jkauppa

2 жыл бұрын

many randomized starts to solve most of the issues, ie, grid starting location search

@Jkauppa

2 жыл бұрын

if you have a higher value somewhere and lower somewhere, you are guaranteed to have the middle values between through some route (1d or 2d, or multi dimensional)

@Jkauppa

2 жыл бұрын

so you dont get stuck in the flat derivatives, you can use the point-to-point derivatives, roughly, to get a next useful point

@Jkauppa

2 жыл бұрын

the grid search with some binary search narrowing criteria, without knowing anything about gradients or derivatives

@Jkauppa

2 жыл бұрын

if you have gradient or derivative of zero, then you either have fully horizontal plane or line, or are guaranteed to have either lower or higher values left and right