MMAD
A formula-driven framework for maximizing target functions via the minorization-maximization (MM) algorithm. The package represents the target as a symbolic expression tree, infers its curvature via disciplined-convex-programming rules, and constructs a separable surrogate at each iterate using only Jensen's inequality and the supporting hyperplane. The driver maximizes the surrogate via block-coordinate Newton with line search, falling back to a multivariate step on any non-separable residue. A formula interface accepts standard R expressions (including 'sum()' reductions and 'X %*% theta' design-matrix products) so statistical models such as Poisson regression can be written in one line.
This page was built for software: MMAD
