Optimization
By Vlad Feinberg - October 27, 2020
Following our previous post on Quasi-Newton methods, having established the Wolfe line search algorithm and how it guarantees convergence under directional Hessian approximation, we move on to QN methods that create approximations that meet this criterion and thus enjoy global and local convergence properties.
Materials
Why BFGS?
Nuggets
Raw Notes
If you like applying these kinds of methods practical ML problems, join our team.