Maximum Likelihood Methods in Unknown Latent Variables (Maximum Likelihood Methods in Unknown Latent Variables)

From Algorithm Wiki
Jump to navigation Jump to search

Description

In this problem, the goal is to compute maximum-likelihood estimates when the observations can be viewed as incomplete data.

Parameters

$n$: number of observations in sample

$r$: number of parameters + latent variables

Table of Algorithms

Name Year Time Space Approximation Factor Model Reference
Expectation-Maximization (EM) algorithm 1977 $O(n^{3})$ $O(n+r)$? Exact Deterministic Time
EM with Quasi-Newton Methods (Jamshidian; Mortaza; Jennrich; Robert I.) 1997 $O(n^{2} \log^{3} n)$ $O(n+r^{2})$? Exact Deterministic Time
Parameter-expanded expectation maximization (PX-EM) 1998 $O(n^{3})$ $O(n+r)$? Exact Deterministic Time
Expectation conditional maximization (ECM) 1993 $O(n^{3})$ $O(n+r)$? Exact Deterministic Time
Expectation conditional maximization either (ECME) (Liu; Chuanhai; Rubin; Donald B) 1994 $O(n^{3})$ $O(n+r)$? Exact Deterministic Time
α-EM Algorithm 2003 $O(n^{3})$ $O(n+r)$? Exact Deterministic Time
Shaban; Amirreza; Mehrdad; Farajtabar 2015 $O(n^{2} \log^{2} n)$ $O(kd+d^{3})$?? Exact Deterministic Time
alpha-HMM (Matsuyama, Yasuo) 2011 $O(n^{2} \log^{2} n)$ Exact Deterministic Time

Time Complexity Graph

Maximum Likelihood Methods in Unknown Latent Variables - Time.png