Decision making in classification problems
View Patent ↗A method of classifying samples to one of a number of predetermined classes involves using a number of class models or classifiers to form order statistic for each classifier. A linear combination of the order statistic (L-statistic) is calculated to determine the confidence of that particular classifier, both in general and for that particular sample. Relative weights are then derived from these confidences, and used to calculate a weighted summation across all classifiers for each class of the likelihoods that a sample belongs to that class. The sample is classified in the class which has the associated weighted summation which is greatest in value.
1. A method for determining a manner of classifying data samples in one of a number of predetermined classes comprising first and second classes, said method comprising:
associating a plurality of data classifiers in a decision fusion application comprising said data samples, wherein said data classifiers indicate a manner of classifying said data sample in said one of a number of fast classes;
computing sample confidence values for each data sample;
determining an overall confidence value for said first classes using said sample confidence values;
assigning a weight value for each of said plurality of data classifiers as a function of said overall confidence value and said sample confidence values;
classifying each said data sample in a second class by calculating a combined log-likelihood value for each second class, wherein said log-likelihood comprises a summation of likelihoods of said plurality of data classifiers weighted by said weight value; and
classifying a calculated second class as a correct class for a particular data sample by selecting a particular second class with a highest calculated combined log-likelihood value.
2. The method of claim 1 , wherein said sample confidence values comprises a linear combination of an order statistic.
3. The method of claim 2 , wherein said linear combination comprises a log-likelihood of respective predetermined classes for said plurality of data classifiers corresponding to said data sample.
4. The mood of claim 2 , wherein said plurality of data classifiers comprise audio data classifiers and video data classifiers.
5. The method of claim 3 , wherein said sample confidence values comprises said log-likelihood of respective predetermined classes for said plurality of data classifiers corresponding to said data sample, and said overall confidence value comprises a mean of said sample confidence values over a plurality of said data samples.
6. The method of claim 5 , wherein said overall confidence value is successively updated with said data sample confidence values of each said data sample.
7. A program storage device readable by computer, tangibly embodying a program of instructions executable by said computer to perform a method for determining a manner of classifying data samples in one of a number of predetermined classes comprising first and second classes, said method comprising:
associating a plurality of data classifiers in a decision fusion application comprising said data samples, wherein said data classifiers indicate a manner of classifying said data sample in said one of a number of first classes;
computing sample confidence values for each data sample;
determining an overall confidence value for said first classes using said sample confidence values;
assigning a weight value for each of said plurality of data classifiers as a function of said overall confidence value and said sample confidence values;
classifying each said data sample in a second class by calculating a combined log-likelihood value for each second class, wherein said log-likelihood comprises a summation of likelihoods of said plurality of data classifiers weighted by said weight value; and
classifying calculated second class as a correct class for a particular data sample by selecting a particular second class with a highest calculated combined log-likelihood value.
8. The program storage device of claim 7 , wherein said sample confidence values comprises a liner combination of an order statistic.
9. The program storage device of claim 8 , wherein said linear combination comprises a log-likelihood of respective predetermined classes for said plurality of data classifiers corresponding to said data sample.
10. The program storage device of claim 8 , wherein said plurality of data classifiers comprise audio data classifiers and video data classifiers.
11. The program storage device of claim 9 , wherein said sample confidence values comprises said log-likelihood of respective predetermined classes for said plurality of data classifiers corresponding to said data sample, and said overall confidence value comprises a mean of said sample confidence values over a plurality of said data samples.
12. The program storage device of claim 11 , wherein said overall confidence value is successively updated with said data sample confidence values of each said data sample.
13. An apparatus for determining a manner of classifying data samples in one of a number of predetermined classes comprising first and second classes, said apparatus comprising:
means for associating a plurality of data classifiers in a decision fusion application comprising said samples, wherein said data classifiers indicate a manner of classifying said data sample in said one of a number of first classes;
means for computing sample confidence values for each data sample;
means for determining an overall confidence value for said first classes using said sample confidence values;
means for assigning a weight value for each of said plurality of data classifiers as a function of said overall confidence value and said sample confidence values;
means for classifying each said data sample in a second class by calculating a combined log-likelihood value for each second class, wherein said log-likelihood comprises a summation of likelihoods of said plurality of data classifiers weighted by said weight value; and
means for classifying a calculated second class as a correct class for a particular data sample by selecting a particular second class with a highest calculated combined log-likelihood value.
14. The method of claim 1 , wherein said decision fusion application comprises an audiovisual speech recognition application.
15. The method of claim 1 , further comprising determining a relative confidence level relating to an accuracy of said plurality of data classifiers for each data sample in said decision fusion application based on said sample confidence values and said overall confidence value.
16. The program storage device of claim 7 , wherein said decision fusion application comprises an audiovisual speech recognition application.
17. The program storage device of claim 7 , wherein said method further comprises determining a relative confidence level relating to an accuracy of said plurality of data classifiers for each data sample in said decision fusion application based on said sample confidence values and said overall confidence value.