Method for training an adaptive statistical classifier to discriminate against inproper patterns
First Claim
1. A method for training a statistical classifier to estimate the probabilities that input patterns are associated with each of a predetermined set of classes, comprising the steps of:
- selecting a first, positive set of training patterns each associated with a class in said set;
selecting a second, negative set of training patterns not associated with any class in said set;
combining said first set and said second sets into a training set;
processing training patterns in said training set through a process comprising the following steps for each training pattern that is processed;
computing a set of target values corresponding to each class in said set of classes, such that;
for a training pattern from said first, positive set, the target value corresponding to its associated class is substantially equal to a first predetermined value, and the other target values of said set of target values are all substantially equal to a second predetermined value that is substantially different from said first predetermined value, and;
for a training pattern from said second set, negative set, all of said target values are substantially equal to said second predetermined value; and
providing said training pattern and said set of target values to a statistical classifier and training said classifier in accordance therewith.
2 Assignments
0 Petitions
Accused Products
Abstract
A statistical classifier that can be used for pattern recognition is trained to recognize negative, or improper patterns as well as proper patterns that are positively associated with desired output classes. A set of training samples includes both the negative and positive patterns, and target output values for the negative patterns are set so that no recognized class is indicated. The negative patterns are selected for training with less frequency than the positive patterns, and their effect on training is also modified, so that training is focused more heavily on positive patterns.
46 Citations
13 Claims
-
1. A method for training a statistical classifier to estimate the probabilities that input patterns are associated with each of a predetermined set of classes, comprising the steps of:
-
selecting a first, positive set of training patterns each associated with a class in said set; selecting a second, negative set of training patterns not associated with any class in said set; combining said first set and said second sets into a training set; processing training patterns in said training set through a process comprising the following steps for each training pattern that is processed; computing a set of target values corresponding to each class in said set of classes, such that; for a training pattern from said first, positive set, the target value corresponding to its associated class is substantially equal to a first predetermined value, and the other target values of said set of target values are all substantially equal to a second predetermined value that is substantially different from said first predetermined value, and; for a training pattern from said second set, negative set, all of said target values are substantially equal to said second predetermined value; and providing said training pattern and said set of target values to a statistical classifier and training said classifier in accordance therewith. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10)
-
-
11. A method for training a statistical classifier to estimate the probability that an input pattern belongs to a predetermined class, comprising the steps of:
-
defining a first, positive set of training patterns, each of which is labeled as belonging to a respective one of a plurality of predetermined classes; defining a second, negative set of training patterns which are not associated with any of said predetermined classes; processing said training patterns in the statistical classifier to produce output values and associating a target value with each training pattern, wherein the target value associated with a training pattern from said first, positive set designates the class to which said pattern belongs, and the target value associated with a training pattern from said second, negative set designates none of said classes; determining an error value for each processed pattern, based upon the difference between the target value associated with the pattern and the output value produced by the classifier in response to the processed pattern; and adjusting operating parameters of said classifier in accordance with said error value. - View Dependent Claims (12, 13)
-
Specification