Methods And Logic For Autonomous Generation Of Ensemble Classifiers, And Systems Incorporating Ensemble Classifiers
First Claim
1. A method for generating an ensemble classifier, the method comprising:
- transforming, automatically with a processor, multidimensional training data into a plurality of response planes according to a plurality of recognition algorithms, wherein each of the response planes comprise a set of confidence scores;
transforming, automatically with the processor, the response planes into a plurality of binary response planes, wherein each of the binary response planes comprise a set of binary scores corresponding to one of the confidence scores;
transforming, automatically with the processor, a first combination of the binary response planes into a first set of diversity metrics according to a first diversity measure;
transforming, automatically with the processor, a second combination of the binary response planes into a second set of diversity metrics according to a second diversity measure;
selecting a first metric from the first set of diversity metrics;
selecting a second metric from the second set of diversity metrics;
generating, automatically with the processor, a predicted performance of a child combination of the recognition algorithms corresponding to the first combination and the second combination, wherein the predicted performance is based at least in part upon the first metric and the second metric;
selecting parent recognition algorithms from the recognition algorithms based at least in part upon the predicted performance; and
generating the ensemble classifier, wherein the ensemble classifier comprises the parent recognition algorithms.
1 Assignment
0 Petitions
Accused Products
Abstract
In one embodiment, a method for generating an ensemble classifier may include transforming multidimensional training data into a plurality of response planes. Each of the response planes includes a set of confidence scores. The response planes are transformed into a plurality of binary response planes. Each of the binary response planes include a set of binary scores corresponding to one of the confidence scores. Combinations of the binary response planes are transformed into sets of diversity metrics according to a diversity measure. A metric is selected from the sets of diversity metrics. A predicted performance of a child combination of the recognition algorithms corresponding to the combinations is generated. The predicted performance is based at least in part upon the metrics. Parent recognition algorithms are selected from the recognition algorithms based at least in part upon the predicted performance. The ensemble classifier is generated and includes the parent recognition algorithms.
-
Citations
20 Claims
-
1. A method for generating an ensemble classifier, the method comprising:
-
transforming, automatically with a processor, multidimensional training data into a plurality of response planes according to a plurality of recognition algorithms, wherein each of the response planes comprise a set of confidence scores; transforming, automatically with the processor, the response planes into a plurality of binary response planes, wherein each of the binary response planes comprise a set of binary scores corresponding to one of the confidence scores; transforming, automatically with the processor, a first combination of the binary response planes into a first set of diversity metrics according to a first diversity measure; transforming, automatically with the processor, a second combination of the binary response planes into a second set of diversity metrics according to a second diversity measure; selecting a first metric from the first set of diversity metrics; selecting a second metric from the second set of diversity metrics; generating, automatically with the processor, a predicted performance of a child combination of the recognition algorithms corresponding to the first combination and the second combination, wherein the predicted performance is based at least in part upon the first metric and the second metric; selecting parent recognition algorithms from the recognition algorithms based at least in part upon the predicted performance; and generating the ensemble classifier, wherein the ensemble classifier comprises the parent recognition algorithms. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17)
-
-
18. A logic for generating an ensemble classifier stored on a machine readable medium, the logic comprising:
-
a receiver-operator characteristic algorithm for generating a receiver-operator characteristic curve for a first two-dimensional image detection algorithm; an area algorithm for generating a first area under a receiver-operator characteristic curve for the first two-dimensional image detection algorithm; a predictive model to estimate a second area under a receiver-operator characteristic curve the ensemble classifier from a first diversity metric of the first two-dimensional image detection algorithm and a second diversity metric of a second two-dimensional image detection algorithm; and a weighting algorithm for combining the first two-dimensional image detection algorithm and the second two-dimensional image detection algorithm, wherein outputs of the first two-dimensional image detection algorithm and the second two-dimensional image detection algorithm are exponentially weighted. - View Dependent Claims (19)
-
-
20. A detection system comprising:
-
a sensor that receives analysis data; a machine readable medium communicably coupled to the sensor; an ensemble classifier stored on the machine readable medium, wherein the ensemble classifier comprises parent recognition algorithms having a predicted performance of a low dependency, such that the predicted performance is a function of a minimum double-fault measure of a recognition algorithm; and a processor communicably coupled to the sensor, wherein the processor executes the ensemble classifier to detect a target within the analysis data.
-
Specification