Abstract
Information divergence functions play a critical role in statistics and information theory. In this paper we show that a nonparametric-divergence measure can be used to provide improvedboundsontheminimumbinaryclassificationprobabilityof error for the case when the training and test data are drawn from the same distribution and for the case where there exists some mismatch between training and test distributions. We confirm these theoretical results by designing feature selection algorithms using the criteria from these bounds and by evaluating the algorithms on a series of pathological speech classification tasks.
| Original language | English (US) |
|---|---|
| Article number | 7254229 |
| Pages (from-to) | 580-591 |
| Number of pages | 12 |
| Journal | IEEE Transactions on Signal Processing |
| Volume | 64 |
| Issue number | 3 |
| DOIs | |
| State | Published - Feb 1 2016 |
Fingerprint
Keywords
- Bayes error rate
- Classification
- Divergence measures
- Domain adaptation
- Nonparametric divergence estimator
ASJC Scopus subject areas
- Signal Processing
- Electrical and Electronic Engineering
Cite this
Empirically estimable classification bounds based on a nonparametric divergence measure. / Berisha, Visar; Wisler, Alan; Hero, Alfred O.; Spanias, Andreas.
In: IEEE Transactions on Signal Processing, Vol. 64, No. 3, 7254229, 01.02.2016, p. 580-591.Research output: Contribution to journal › Article
}
TY - JOUR
T1 - Empirically estimable classification bounds based on a nonparametric divergence measure
AU - Berisha, Visar
AU - Wisler, Alan
AU - Hero, Alfred O.
AU - Spanias, Andreas
PY - 2016/2/1
Y1 - 2016/2/1
N2 - Information divergence functions play a critical role in statistics and information theory. In this paper we show that a nonparametric-divergence measure can be used to provide improvedboundsontheminimumbinaryclassificationprobabilityof error for the case when the training and test data are drawn from the same distribution and for the case where there exists some mismatch between training and test distributions. We confirm these theoretical results by designing feature selection algorithms using the criteria from these bounds and by evaluating the algorithms on a series of pathological speech classification tasks.
AB - Information divergence functions play a critical role in statistics and information theory. In this paper we show that a nonparametric-divergence measure can be used to provide improvedboundsontheminimumbinaryclassificationprobabilityof error for the case when the training and test data are drawn from the same distribution and for the case where there exists some mismatch between training and test distributions. We confirm these theoretical results by designing feature selection algorithms using the criteria from these bounds and by evaluating the algorithms on a series of pathological speech classification tasks.
KW - Bayes error rate
KW - Classification
KW - Divergence measures
KW - Domain adaptation
KW - Nonparametric divergence estimator
UR - http://www.scopus.com/inward/record.url?scp=85009355217&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85009355217&partnerID=8YFLogxK
U2 - 10.1109/TSP.2015.2477805
DO - 10.1109/TSP.2015.2477805
M3 - Article
AN - SCOPUS:85009355217
VL - 64
SP - 580
EP - 591
JO - IEEE Transactions on Signal Processing
JF - IEEE Transactions on Signal Processing
SN - 1053-587X
IS - 3
M1 - 7254229
ER -