Please use this identifier to cite or link to this item:
http://dspace.mediu.edu.my:8181/xmlui/handle/1721.1/6620Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.creator | Jordan, Michael I. | - |
| dc.creator | Xu, Lei | - |
| dc.date | 2004-10-08T20:34:35Z | - |
| dc.date | 2004-10-08T20:34:35Z | - |
| dc.date | 1993-11-01 | - |
| dc.date.accessioned | 2013-10-09T02:46:10Z | - |
| dc.date.available | 2013-10-09T02:46:10Z | - |
| dc.date.issued | 2013-10-09 | - |
| dc.identifier | AIM-1458 | - |
| dc.identifier | http://hdl.handle.net/1721.1/6620 | - |
| dc.identifier.uri | http://koha.mediu.edu.my:8181/xmlui/handle/1721 | - |
| dc.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. | - |
| dc.format | 245749 bytes | - |
| dc.format | 829871 bytes | - |
| dc.format | application/octet-stream | - |
| dc.format | application/pdf | - |
| dc.language | en_US | - |
| dc.relation | AIM-1458 | - |
| dc.title | Convergence Results for the EM Approach to Mixtures of Experts Architectures | - |
| Appears in Collections: | MIT Items | |
Files in This Item:
There are no files associated with this item.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.