Métodos de optimización convexa para resolver problemas en aprendizaje automático

Abstract

Nowadays, the learning methods developed to solve optimization problems turn out to have strange behavior when we work analyzing data of a large dimension, mainly developed in a statistical context. Many of these methods have become computationally hard and slow. Right, the bene ts of methods optimization only come when the problem is kwnown ahead of time to be convex. Methods, such as gradient descent or geometric descent, are capable of dealing with these problems e ciently, thus obtaining great results. Throughout this work, we will also expose and develop other methods derived from machine learning, thus concluding with the bootstrap model, which has now became relevant.