Home People Research Publications Demos
         
News Jobs Prospective
Students
About Internal

Cost Sensitive Learning

Classification problems such as fraud detection, medical diagnosis, or object detection in computer vision, are naturally cost sensitive. In these problems the cost of missing a target is much higher than that of a false-positive, and classifiers that are optimal under symmetric costs (such as the popular zero-one loss) tend to under perform. The design of optimal classifiers with respect to losses that weigh certain types of errors more heavily than others is denoted as cost-sensitive learning.

Algorithms:
Cost Sensitive Boosting
We derive the cost sensitive AdaBoost, RealBoost and LogitBoost algorithms and utilize them for computer vision and medical diagnosis applications with state of the art results.
    

Cost Sensitive SVM
We derive the cost sensitive SVM algorithm and utilize it for fraud detection, business decision making and medical diagnosis applications with state of the art results.
    


Publications: Cost-Sensitive Boosting.
Hamed Masnadi-Shirazi and Nuno Vasconcelos
IEEE Trans. Pattern Analysis and Machine Intelligence,
2010.
[pdf]

Risk minimization, probability elicitation, and cost-sensitive SVMs
Hamed Masnadi-Shirazi and Nuno Vasconcelos.
International Conference on Machine Learning (ICML), 2010.
(acceptance rate 20%)
[pdf]

High Detection-rate Cascades for Real-Time Object Detection.
Hamed Masnadi-Shirazi and Nuno Vasconcelos
Proceedings of IEEE International Conference on Computer Vision (ICCV) ,
Rio de Janeiro, Brazil, 2007.
IEEE, [pdf]

  Asymmetric Boosting
Hamed Masnadi-Shirazi and Nuno Vasconcelos
Proceedings of International Conference on Machine Learning (ICML),
Corvallis, OR, May 2007.
[pdf]
Contact: Nuno Vasconcelos, Hamed Masnadi-Shirazi

 



Copyright @ 2007 www.svcl.ucsd.edu