I am searching for a function to show the "zig-zag" pattern of the steepest decent, I want to show that SGD converges after less iterations.
Which functions should look at?
I am searching for a function to show the "zig-zag" pattern of the steepest decent, I want to show that SGD converges after less iterations.
Which functions should look at?
Steepest Descent (gradient descent with line search) exhibits zigzag behavior on the Rosenbrock Banana function.
$(x-1)^2 + 100(y-x^2)^2$ using the starting point (-1.2,1). Lots of zigzags to get to the optimum at (1,1). You can make it zigzag even more by increasing 100 to a larger number.