Thesiss

Many machine learning problems can be formulated as finite-sum nonconvex optimization problems, where each objective function component is the loss of a training sample. The computational complexity to solve such optimization problems scales with the total number of training sam…