Sunday, Jul 31, 2011 (IJCNN 2011)
San Jose, CA, 1:30 - 3:30 pm
Monday, Aug 8, 2011 (AAAI 2011)
San Francisco, CA, 2:00 - 6:00 pm
The Conformal Predictions framework is a recent development in machine learning to associate reliable measures of confidence with results in classification and regression. This theory is based on the relationship derived between transductive inference and the randomness deficiency of an i.i.d. (identically independently distributed) sequence of data instances. One of the desirable features of this framework is the calibration of the obtained confidence values in an online setting. While probability/confidence values generated by existing approaches can often be unreliable and difficult to interpret, the theory behind the CP framework guarantees that the confidence values obtained using this transductive inference framework manifest as the actual error frequencies in the online setting i.e. they are well-calibrated. Further, this framework can be applied across all existing classification and regression methods (such as neural networks, Support Vector Machines, k-Nearest Neig! hbors, ridge regression, etc), thus making it a very generalizable approach.
Over the last few years, there has been a growing interest in applying this framework to real-world problems such as clinical decision support, medical diagnosis, sea surveillance, network traffic classification, and face recognition. The promising results have generated in further extensions of the framework to problem settings beyond just classification or regression. The framework has now been extended towards newer settings such as active learning, model selection, feature selection, change detection, outlier detection, and anomaly detection.
The key objectives of this tutorial are:
- to expose the audience to the basic theory of the Conformal Predictions framework
- to demonstrate examples of how the framework can be applied in real-world problems (including code simulations), and
- to provide sample adaptations of the framework to related machine learning problems such as active learning, transfer learning, anomaly detection and model selection, illustrating the potential of the framework in machine learning applications.
Vineeth N Balasubramanian, Arizona State University
Shen-Shyang Ho, University of Maryland
Sethuraman Panchanathan, Arizona State University
Vladimir Vovk, Royal Holloway University of London
Please see the website (http://www.public.asu.edu/~