Computational Mathematics for Learning and Data Analysis - AA 2024/25

Better nondifferentiable optimization methods in theory (and sometimes in practice): smoothing methods. The pitfalls (you need to be able to change the function), how it goes (not fast). Better nondifferentiable optimization methods in practice (although not in theory): the cutting plane method (not better either way) and its really better variants (Bundle methods). A fleeting glimpse to the theory. MATLAB implementation of the proximal bundle method, its behaviour: accruing information does work (much faster convergence speed and effective stopping criterion), at a cost.