Abstract
We study the problem of on-line classification in which the prediction algorithm is given a “confidence level” 1-δ and is required to output as its prediction a range of labels (intuitively, those labels deemed compatible with the available data at the level δ) rather than just one label; as usual, the examples are assumed to be generated independently from the same probability distribution P. The prediction algorithm is said to be “well-calibrated” for P and δ if the long-run relative frequency of errors does not exceed δ almost surely w.r. to P. For well-calibrated algorithms we take the number of “uncertain” predictions (i.e., those containing more than one label) as the principal measure of predictive performance. The main result of this paper is the construction of a prediction algorithm which, for any (unknown) P and any δ: (a) makes errors independently and with probability δ at every trial (in particular, is well-calibrated for P and δ); (b) makes in the long run no more uncertain predictions than any other prediction algorithm that is well-calibrated for P and δ; (c) processes example n in time O(logn).
This is a preview of subscription content, log in via an institution.
Buying options
Tax calculation will be finalised at checkout
Purchases are for personal use only
Learn about institutional subscriptionsPreview
Unable to display preview. Download preview PDF.
References
Cormen, T.H., Leiserson, C.E., Rivest, R.L., Stein, C.: Introduction to Algorithms, 2nd edn. MIT Press, Cambridge (2001)
Cox, D.R., Hinkley, D.V.: Theoretical Statistics. Chapman and Hall, London (1974)
Devroye, L., Gyórfi, L., Lugosi, G.: A Probabilistic Theory of Pattern Recognition. Springer, New York (1996)
Freund, Y., Mansour, Y., Schapire, R.E.: Generalization bounds for averaged classifiers. To appear in Annals of Statistics (April 2002)
Rivest, R.L., Sloan, R.: Learning complicated concepts reliably and usefully. In: Proceedings of the 1st Annual Conference on Computational Learning Theory, San Mateo, CA, pp. 69–79. Morgan Kaufmann, San Francisco (1988)
Saunders, C., Gammerman, A., Vovk, V.: Transduction with confidence and credibility. In: Proceedings of the 16th International Joint Conference on Artificial Intelligence, pp. 722–726 (1999)
Shiryaev, A.N.: Probability, 2nd edn. Springer, New York (1996)
Vovk, V.: Asymptotic optimality of Transductive Confidence Machine. In: Cesa-Bianchi, N., Numao, M., Reischuk, R. (eds.) ALT 2002. LNCS (LNAI), vol. 2533, pp. 336–350. Springer, Heidelberg (2002)
Vovk, V.: On-line Confidence Machines are well-calibrated. In: Proceedings of the 43rd Annual Symposium on Foundations of Computer Science, pp. 187–196. IEEE Computer Society, Los Alamitos (2002)
Vovk, V.: Universal well-calibrated algorithm for on-line classification, Online Compression Modelling project Working Paper #3 (November 2002), http://vovk.net/kp
Vovk, V., Gammerman, A., Saunders, C.: Machine-learning applications of algorithmic randomness. In: Proceedings of the 16th International Conference on Machine Learning, San Francisco, CA, pp. 444–453. Morgan Kaufmann, San Francisco (1999)
Author information
Authors and Affiliations
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2003 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Vovk, V. (2003). Universal Well-Calibrated Algorithm for On-Line Classification. In: Schölkopf, B., Warmuth, M.K. (eds) Learning Theory and Kernel Machines. Lecture Notes in Computer Science(), vol 2777. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-540-45167-9_27
Download citation
DOI: https://doi.org/10.1007/978-3-540-45167-9_27
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-40720-1
Online ISBN: 978-3-540-45167-9
eBook Packages: Springer Book Archive