Year: 2019
Type: article
Abstract: The need for optimization is almost everywhere today, especially with the rise of deep learning approaches to AI that require searching through millions of neural network weights and at least dozens of architectural hyperparameters.Optimization problems consist of finding a maximum or minimum value of a objective function (in most practical cases, subject to a set of constraints).Depending on the inherent complexity and characteristics of the function to be optimized, there is a diverse armamentarium of ways to attack optimization problems.Methods including Linear and Nonlinear programming, Dynamic programming, Calculus of Variations, and Optimal Control have long been successfully used on smaller, traditional problems and can in some cases be completely dominant in a given area of application.
DOI: 10.21105/joss.01839
Authors: Georgios Detorakis, Andrew Burton
Institution: University of California, Irvine
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W2992284548