DocumentCode :
2621758
Title :
Classification rules in the unknown mixture parameter case: relative value of labeled and unlabeled samples
Author :
Castelli, Vittorio ; Cover, Thomas M.
Author_Institution :
Dept. of Electr. Eng., Stanford Univ., CA, USA
fYear :
1994
fDate :
27 Jun-1 Jul 1994
Firstpage :
111
Abstract :
We investigate the relative value of labeled and unlabeled samples in constructing classification rules. We observe a training set Q composed of l labeled and u unlabeled samples coming from two classes. Let sample from class 1 be distributed according to f1(·), samples from class 2 according to f2(·), and let η be the probability that a sample is in class 1. Assume that f1(·) and f2(·) are known and that η is unknown. We want to classify a new sample X0. The relative value of labeled and unlabeled observations in reducing the probability of error is equal to It(η)/Iu(η), the ratio of the Fisher information of the labeled and unlabeled samples. Moreover labeled samples are not necessary in order to construct a decision rule. However, if f1(·) and f2(·) are given, but it is not known whether observations from class 1 are distributed according to f1(·) or according to f2 (·), then the labeled samples are necessary and exponentially more valuable than unlabeled samples
Keywords :
error statistics; parameter estimation; probability; signal sampling; Fisher information; classification rules; decision rule; error probability; labeled samples; training set; unknown mixture parameter; unlabeled samples; Computer aided software engineering; Information systems; Labeling; Laboratories; Random variables; Testing;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Information Theory, 1994. Proceedings., 1994 IEEE International Symposium on
Conference_Location :
Trondheim
Print_ISBN :
0-7803-2015-8
Type :
conf
DOI :
10.1109/ISIT.1994.394877
Filename :
394877
Link To Document :
بازگشت