Results 1 
7 of
7
The ClassSpecific Classifier: Avoiding the Curse of Dimensionality
 IEEE Aerosp. Electron. Syst. Mag
, 2002
"... this article is to introduce the reader to the basic principles of classification with classspecific features. It is written both for readers interested in only the basic concepts as well as those interested in getting started in applying the method. For indepth coverage, the reader is referred to ..."
Abstract

Cited by 12 (3 self)
 Add to MetaCart
(Show Context)
this article is to introduce the reader to the basic principles of classification with classspecific features. It is written both for readers interested in only the basic concepts as well as those interested in getting started in applying the method. For indepth coverage, the reader is referred to a more detailed article [1]
Conditional Model Order Estimation
 IEEE Transactions on Signal Processing
, 2001
"... Abstract—A new approach to model order selection is proposed. Based on the theory of sufficient statistics, the method does not require any prior knowledge of the model parameters. It is able to discriminate between models by basing the decision on the part of the data that is independent of the mod ..."
Abstract

Cited by 9 (0 self)
 Add to MetaCart
(Show Context)
Abstract—A new approach to model order selection is proposed. Based on the theory of sufficient statistics, the method does not require any prior knowledge of the model parameters. It is able to discriminate between models by basing the decision on the part of the data that is independent of the model parameters. This is accomplished conceptually by transforming the data into a sufficient statistic and an ancillary statistic with respect to the model parameters. It is the probability density function of the ancillary statistic when adjusted for its dimensionality that is used to estimate the order. Furthermore, the rule is directly tied to the goal of minimizing the probability of error and does not employ any asymptotic approximations. The estimator can be shown to be consistent and, via computer simulation, is found to outperform the minimum description length estimator. Index Terms—Adaptive signal detection, modeling, spectral analysis, speech analysis. I.
classification, and the classspecific feature theorem
 IEEE Trans. Inform. Theory
, 2000
"... Abstract—A new proof of the classspecific feature theorem is given. The proof makes use of the observed data as opposed to the set of sufficient statistics as in the original formulation. We prove the theorem for the classical case, in which the parameter vector is deterministic and known, as well ..."
Abstract

Cited by 4 (0 self)
 Add to MetaCart
Abstract—A new proof of the classspecific feature theorem is given. The proof makes use of the observed data as opposed to the set of sufficient statistics as in the original formulation. We prove the theorem for the classical case, in which the parameter vector is deterministic and known, as well as for the Bayesian case, in which the parameter vector is modeled as a random vector with known prior probability density function. The essence of the theorem is that with a suitable normalization the probability density function of the sufficient statistic for each probability density function family can be used for optimal classification. One need not have knowledge of the probability density functions of the data under each hypothesis. Index Terms—Bayes procedures, data models, information theory, pattern recognition, signal detection.
Model Selection Within a Bayesian Approach to Extraction of Walker Motion
"... Extracting articulated motion of walking people in image sequences remains a challenge, particularly when we take into account the changes caused by carried objects or the severe motion occlusions by clothing (e.g., a long skirt, a trench coat, etc). In this paper, we propose a Bayesian framework ca ..."
Abstract

Cited by 1 (0 self)
 Add to MetaCart
(Show Context)
Extracting articulated motion of walking people in image sequences remains a challenge, particularly when we take into account the changes caused by carried objects or the severe motion occlusions by clothing (e.g., a long skirt, a trench coat, etc). In this paper, we propose a Bayesian framework capable of handling such uncertainties by exploiting our strong prior knowledge of how humans walk. In this work, the strong prior is built from a simple articulated model, which can be easily modified to cater for situations such as walkers wearing clothing that obscures the limbs. A model selection process is built into the framework to determine the body configuration of the walker in the given sequence automatically. The statistics of the parameters describing a basic walker are learned from data and the Bayesian framework then allows us to ‘bootstrap ’ to accurate motion extraction on the images of walkers with extra body configurations. We demonstrate our approach on the data of walkers with rucksacks, skirts and trench coats. Results are quantified in terms of average pixel error between automatically extracted body points and corresponding points labelled by hand. 1.
IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART B: CYBERNETICS 1 Joint Segmentation and Classification of Time Series Using ClassSpecific Features
"... Abstract—We present an approach for the joint segmentation and classification of a time series. The segmentation is on the basis of a menu of possible statistical models: each of these must be describable in terms of a sufficient statistic, but there is no need for these sufficient statistics to be ..."
Abstract
 Add to MetaCart
(Show Context)
Abstract—We present an approach for the joint segmentation and classification of a time series. The segmentation is on the basis of a menu of possible statistical models: each of these must be describable in terms of a sufficient statistic, but there is no need for these sufficient statistics to be the same, and these can be as complex (for example, cepstral features or autoregressive coefficients) as fits. All that is needed is the probability density function (PDF) of each sufficient statistic under its own assumed model—presumably this comes from training data, and it is particularly appealing that there is no need at all for a joint statistical characterization of all the statistics. There is similarly no need for an apriori specification of the number of sections, as the approach uses an appropriate penalization of an overzealous segmentation. The scheme has two stages. In stage one, rough segmentations are implemented sequentially using a piecewise generalized likelihood ratio (GLR); in the second stage, the results from the first stage (both forward and backward) are refined. The computational burden is remarkably small, approximately linear with the length of the time series, and the method is nicely accurate in terms both of discovered number of segments and of segmentation accuracy. A hybrid of the approach with one based on Gibbs sampling is also presented; this combination is somewhat slower but considerably more accurate. Index Terms—Classification, classspecific, GLR, order selection, segmentation.
SPEECH ANALYSIS AND COGNITION USING CATEGORYDEPENDENT FEATURES IN A MODEL OF THE CENTRAL AUDITORY SYSTEM Approved by:
, 2006
"... To my mother and father, for their boundless love and patience iii ACKNOWLEDGEMENTS First and foremost, I thank my parents for their unconditional love and support, without which none of the worthwhile achievements in my life would have been possible. I also give my deepest thanks to my advisor, Pr ..."
Abstract
 Add to MetaCart
(Show Context)
To my mother and father, for their boundless love and patience iii ACKNOWLEDGEMENTS First and foremost, I thank my parents for their unconditional love and support, without which none of the worthwhile achievements in my life would have been possible. I also give my deepest thanks to my advisor, Prof. Fred Juang, for his excellent research guidance and training and for teaching me the meaning of hard work and dedication by show of example. He is a true researcher and advisor. I thank the many other faculty members who interacted with me in varying degrees
Real Time Detection of Link Failures in Inter Domain Routing
"... Abstract — Measurements have shown that network path failures occur frequently in the Internet and physical link failures can cause network instability in large scale and severity. Inter domain routing protocols like Border Gateway Protocol (BGP) can take up to 15 minutes to converge after such fail ..."
Abstract
 Add to MetaCart
(Show Context)
Abstract — Measurements have shown that network path failures occur frequently in the Internet and physical link failures can cause network instability in large scale and severity. Inter domain routing protocols like Border Gateway Protocol (BGP) can take up to 15 minutes to converge after such failures and during the convergence period, packets may encounter transient loops, delays and losses [1]. Thus early anomaly detection mechanisms are of great importance. In this paper, we propose a Bayesian approach for time efficient link failure detection using BGP update message traces. The detection is done using an automated mechanism to label, train and classify the network status based on features extracted from BGP traces. In addition to detecting temporal changes in these features, our scheme augments its accuracy by including information on the spatial correlation of the route updates in the decision process. We validate our approach by testing the proposed mechanism on real BGP traces collected during three typical network outage events caused by link failures. I.