Bayesian training of mixture density networks

Lars U. Hjorth; Ian T. Nabney

doi:10.1109/IJCNN.2000.860813

Bayesian training of mixture density networks

Lars U. Hjorth, Ian T. Nabney

Computer Science Research Group

Research output: Chapter in Book/Published conference output › Conference publication

Abstract

Mixture Density Networks (MDNs) are a well-established method for modelling the conditional probability density which is useful for complex multi-valued functions where regression methods (such as MLPs) fail. In this paper we extend earlier research of a regularisation method for a special case of MDNs to the general case using evidence based regularisation and we show how the Hessian of the MDN error function can be evaluated using R-propagation. The method is tested on two data sets and compared with early stopping.

Original language	English
Title of host publication	Proceedings of the IEEE-INNS-ENNS International Joint Conference on Neural Networks, 2000. IJCNN 2000
Place of Publication	Piscataway, NJ, United States
Publisher	IEEE
Pages	455-460
Number of pages	6
Volume	4
ISBN (Print)	9780769506197
DOIs	https://doi.org/10.1109/IJCNN.2000.860813
Publication status	Published - 13 Aug 2000
Event	International Joint Conference on Neural Networks - Como, Italy Duration: 24 Jul 2000 → 27 Jul 2000

Conference

Conference	International Joint Conference on Neural Networks
Abbreviated title	IJCNN 2000
Country/Territory	Italy
City	Como
Period	24/07/00 → 27/07/00

Keywords

Bayes methods
learning
artificial intelligence
neural nets
Bayesian training
MDN error function
mixture density networks
R-propagation
conditional probability density

Access to Document

10.1109/IJCNN.2000.860813

Cite this

@inproceedings{5e7db71514184fb4aa233825b261c32f,

title = "Bayesian training of mixture density networks",

abstract = "Mixture Density Networks (MDNs) are a well-established method for modelling the conditional probability density which is useful for complex multi-valued functions where regression methods (such as MLPs) fail. In this paper we extend earlier research of a regularisation method for a special case of MDNs to the general case using evidence based regularisation and we show how the Hessian of the MDN error function can be evaluated using R-propagation. The method is tested on two data sets and compared with early stopping.",

keywords = "Bayes methods, learning, artificial intelligence, neural nets, Bayesian training, MDN error function, mixture density networks, R-propagation, conditional probability density",

author = "Hjorth, {Lars U.} and Nabney, {Ian T.}",

year = "2000",

month = aug,

day = "13",

doi = "10.1109/IJCNN.2000.860813",

language = "English",

isbn = "9780769506197",

volume = "4",

pages = "455--460",

booktitle = "Proceedings of the IEEE-INNS-ENNS International Joint Conference on Neural Networks, 2000. IJCNN 2000",

publisher = "IEEE",

address = "United States",

note = "International Joint Conference on Neural Networks, IJCNN 2000 ; Conference date: 24-07-2000 Through 27-07-2000",

}

TY - GEN

T1 - Bayesian training of mixture density networks

AU - Hjorth, Lars U.

AU - Nabney, Ian T.

PY - 2000/8/13

Y1 - 2000/8/13

N2 - Mixture Density Networks (MDNs) are a well-established method for modelling the conditional probability density which is useful for complex multi-valued functions where regression methods (such as MLPs) fail. In this paper we extend earlier research of a regularisation method for a special case of MDNs to the general case using evidence based regularisation and we show how the Hessian of the MDN error function can be evaluated using R-propagation. The method is tested on two data sets and compared with early stopping.

AB - Mixture Density Networks (MDNs) are a well-established method for modelling the conditional probability density which is useful for complex multi-valued functions where regression methods (such as MLPs) fail. In this paper we extend earlier research of a regularisation method for a special case of MDNs to the general case using evidence based regularisation and we show how the Hessian of the MDN error function can be evaluated using R-propagation. The method is tested on two data sets and compared with early stopping.

KW - Bayes methods

KW - learning

KW - artificial intelligence

KW - neural nets

KW - Bayesian training

KW - MDN error function

KW - mixture density networks

KW - R-propagation

KW - conditional probability density

UR - http://www.scopus.com/inward/record.url?scp=0033685656&partnerID=8YFLogxK

UR - http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=860813

U2 - 10.1109/IJCNN.2000.860813

DO - 10.1109/IJCNN.2000.860813

M3 - Conference publication

SN - 9780769506197

VL - 4

SP - 455

EP - 460

BT - Proceedings of the IEEE-INNS-ENNS International Joint Conference on Neural Networks, 2000. IJCNN 2000

PB - IEEE

CY - Piscataway, NJ, United States

T2 - International Joint Conference on Neural Networks

Y2 - 24 July 2000 through 27 July 2000

ER -

Bayesian training of mixture density networks

Abstract

Conference

Keywords

Access to Document

Other files and links

Cite this