Adaptive Least Error Rate Algorithm for Neural Network Classifiers

Chen, S., Mulgrew, B. and Hanzo, L. (2001) Adaptive Least Error Rate Algorithm for Neural Network Classifiers. Neural Networks for Signal Processing, , Falmouth, Mass., United States. 10 - 12 Sep 2001. pp. 223-232 .

Record type: Conference or Workshop Item (Paper)

Abstract

We consider sample-by-sample adaptive training of two-class neural network classifiers. Specific applications that we have in mind are communication channel equalization and code-division multiple-access (CDMA) multiuser detection. Typically, training of such neural network classifiers is done using some stochastic gradient algorithm that tries to minimize the mean square error (MSE). Since the goal should really be minimizing the error probability, the MSE is a "wrong" criterion to use and may lead to a poor performance. We propose a stochastic gradient adaptive minimum error rate (MER) algorithm called the least error rate (LER) for training neural network classifiers.

Other

nnsp01P.ps - Other

Download (5MB)

Text

sqc-bm-lh-Sept01-NNSP01.pdf - Other

Download (290kB)

More information

Published date: September 2001

Additional Information: Presented at 2001 IEEE Workshop on Neural Networks for Signal Processing (Falmouth, MA, USA) , Sept.10-12, 2001 Event Dates: 10-12 September 2001 Organisation: IEEE Signal Processing Society

Venue - Dates: Neural Networks for Signal Processing, , Falmouth, Mass., United States, 2001-09-10 - 2001-09-12

Organisations: Southampton Wireless Group

Learn more about Southampton Wireless Group research Learn more about Next Generation Wireless research