Convergence Results for the EM Approach to Mixtures of Experts Architectures
The Expectation-Maximization (EM) algorithm is an iterative approach to maximum likelihood parameter estimation. Jordan and Jacobs (1993) recently proposed an EM algorithm for the mixture of experts architecture of Jacobs, Jordan, Nowlan and Hinton (1991) and the hierarchical mixture of exper...
Main Authors: | , |
---|---|
Language: | en_US |
Published: |
2004
|
Online Access: | http://hdl.handle.net/1721.1/6620 |
_version_ | 1811002331673657344 |
---|---|
author | Jordan, Michael I. Xu, Lei |
author_facet | Jordan, Michael I. Xu, Lei |
author_sort | Jordan, Michael I. |
collection | MIT |
description | The Expectation-Maximization (EM) algorithm is an iterative approach to maximum likelihood parameter estimation. Jordan and Jacobs (1993) recently proposed an EM algorithm for the mixture of experts architecture of Jacobs, Jordan, Nowlan and Hinton (1991) and the hierarchical mixture of experts architecture of Jordan and Jacobs (1992). They showed empirically that the EM algorithm for these architectures yields significantly faster convergence than gradient ascent. In the current paper we provide a theoretical analysis of this algorithm. We show that the algorithm can be regarded as a variable metric algorithm with its searching direction having a positive projection on the gradient of the log likelihood. We also analyze the convergence of the algorithm and provide an explicit expression for the convergence rate. In addition, we describe an acceleration technique that yields a significant speedup in simulation experiments. |
first_indexed | 2024-09-23T15:46:50Z |
id | mit-1721.1/6620 |
institution | Massachusetts Institute of Technology |
language | en_US |
last_indexed | 2024-09-23T15:46:50Z |
publishDate | 2004 |
record_format | dspace |
spelling | mit-1721.1/66202019-04-11T02:52:34Z Convergence Results for the EM Approach to Mixtures of Experts Architectures Jordan, Michael I. Xu, Lei The Expectation-Maximization (EM) algorithm is an iterative approach to maximum likelihood parameter estimation. Jordan and Jacobs (1993) recently proposed an EM algorithm for the mixture of experts architecture of Jacobs, Jordan, Nowlan and Hinton (1991) and the hierarchical mixture of experts architecture of Jordan and Jacobs (1992). They showed empirically that the EM algorithm for these architectures yields significantly faster convergence than gradient ascent. In the current paper we provide a theoretical analysis of this algorithm. We show that the algorithm can be regarded as a variable metric algorithm with its searching direction having a positive projection on the gradient of the log likelihood. We also analyze the convergence of the algorithm and provide an explicit expression for the convergence rate. In addition, we describe an acceleration technique that yields a significant speedup in simulation experiments. 2004-10-08T20:34:35Z 2004-10-08T20:34:35Z 1993-11-01 AIM-1458 http://hdl.handle.net/1721.1/6620 en_US AIM-1458 245749 bytes 829871 bytes application/octet-stream application/pdf application/octet-stream application/pdf |
spellingShingle | Jordan, Michael I. Xu, Lei Convergence Results for the EM Approach to Mixtures of Experts Architectures |
title | Convergence Results for the EM Approach to Mixtures of Experts Architectures |
title_full | Convergence Results for the EM Approach to Mixtures of Experts Architectures |
title_fullStr | Convergence Results for the EM Approach to Mixtures of Experts Architectures |
title_full_unstemmed | Convergence Results for the EM Approach to Mixtures of Experts Architectures |
title_short | Convergence Results for the EM Approach to Mixtures of Experts Architectures |
title_sort | convergence results for the em approach to mixtures of experts architectures |
url | http://hdl.handle.net/1721.1/6620 |
work_keys_str_mv | AT jordanmichaeli convergenceresultsfortheemapproachtomixturesofexpertsarchitectures AT xulei convergenceresultsfortheemapproachtomixturesofexpertsarchitectures |