Monotonicity of quadratic-approximation algorithms (Q757004)

This paper studies a class of quadratic-approximation-based algorithms related to the Newton-Raphson algorithm. The algorithms monotonically increase the objective function and have guaranteed convergence and a linear rate of convergence. The fundamental tool in the analysis is the geometric insight obtained by interpreting quadratic-approximation algorithms as a form of area approximation. Applications of the algorithm are given to calculate maximum likelihood estimation in mixture models, logistic regression, and proportional hazards regression.