Extension of Gradient descent for Convex functions which may not be Differentiable. It uses the Subderivative (and subgradient), instead of derivative/gradient

See here: Analogous of gradient descent for convex functions

Subgradient of hinge loss