Abstract
We present sharp bounds on the risk of the empirical minimization algorithm under mild assumptions on the class. We introduce the notion of isomorphic coordinate projections and show that this leads to a sharper error bound than the best previously known. The quantity which governs this bound on the empirical minimizer is the largest fixed point of the function \(\xi_{n}(r)=\mathbb{E}{\rm sup}\{|\mathbb{E}f-\mathbb{E}_{n}f|:f\in F,\mathbb{E}f=r\}\). We prove that this is the best estimate one can obtain using “structural results”, and that it is possible to estimate the error rate from data. We then prove that the bound on the empirical minimization algorithm can be improved further by a direct analysis, and that the correct error rate is the maximizer of ξ′ n (r) − r, where \(\xi'_{n}(r)=\mathbb{E}{\rm sup}\{\mathbb{E}f-\mathbb{E}_{n}f:f\in F,\mathbb{E}f=r\}\).
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
References
Bartlett, P.L., Boucheron, S., Lugosi, G.: Model selection and error estimation. Machine Learning 48, 85–113 (2002)
Bartlett, P.L., Bousquet, O., Mendelson, S.: Local Rademacher Complexities. (2002), available at http://www.stat.berkeley.edu/~bartlett/publications/recent-pubs.html (submitted)
Bartlett, P.L., Jordan, M.I., McAuliffe, J.D.: Convexity, classification, and risk bounds. Tech. Rep. 638, Dept. of Stat., U.C. Berkeley (2003)
Bartlett, P.L., Mendelson, S.: Empirical minimization. ( 2003), available at http://axiom.anu.edu.au/~shahar (submitted)
Boucheron, S., Lugosi, G., Massart, P.: Concentration inequalities using the entropy method. Ann. of Prob. 31, 1583–1614 (2003)
Bousquet, O.: Concentration Inequalities and Empirical Processes Theory Applied to the Analysis of Learning Algorithms. PhD. Thesis (2002)
Klein, T.: Une inégalité de concentration gauche pour les processus empiriques. C. R. Math. Acad. Sci. Paris 334(6), 501–504 (2002)
Koltchinskii, V.: Rademacher penalties and structural risk minimization. IEEE Trans. on Info. Th. 47(5), 1902–1914 (2001)
Koltchinskii, V.: Local Rademacher Complexities and Oracle Inequalities in Risk Minimization. Tech. Rep, Univ. of New Mexico (August 2003)
Koltchinskii, V., Panchenko, D.: Rademacher processes and bounding the risk of function learning. In: Gine, E., Mason, D., Wellner, J. (eds.) High Dimensional Probability II, pp. 443–459 (2000)
Ledoux, M.: The concentration of measure phenomenon. In: Mathematical Surveys and Monographs, vol. 89 AMS(2001)
Lee, W.S., Bartlett, P.L., Williamson, R.C.: The Importance of Convexity in Learning with Squared Loss. IEEE Trans. on Info Th. 44(5), 1974–1980 (1998)
Lugosi, G. Wegkamp, M.: Complexity regularization via localized random penalties. Ann. of Stat. (2003) (to appear)
Massart, P.: About the constants in Talagrand’s concentration inequality for empirical processes. Ann. of Prob. 28(2), 863–884 (2000)
Massart, P.: Some applications of concentration inequalities to statistics. Ann. de la Faculté des Sciences de Toulouse, IX: 245–303 (2000)
Mendelson, S.: Improving the sample complexity using global data. IEEE Trans. on Info. Th. 48(7), 1977–1991 (2002)
Mendelson, S.: A few notes on Statistical Learning Theory. In: Mendelson, S., Smola, A.J. (eds.) Advanced Lectures on Machine Learning. LNCS (LNAI), vol. 2600, pp. 1–40. Springer, Heidelberg (2003)
Mendelson, S.: Rademacher averages and phase transitions in Glivenko-Cantelli classes. IEEE Transactions on Information Theory 48(1), 251–263 (2002)
Rio, E.: Inégalités de concentration pour les processus empiriques de classes de parties. Probab. Theory Related Fields 119(2), 163–175 (2001)
Talagrand, M.: New concentration inequalities in product spaces. Invent. Math. 126, 505–563 (1996)
Talagrand, M.: Sharper bounds for Gaussian and empirical processes. Ann. of Prob. 22(1), 28–76 (1994)
Author information
Authors and Affiliations
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2004 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Bartlett, P.L., Mendelson, S., Philips, P. (2004). Local Complexities for Empirical Risk Minimization. In: Shawe-Taylor, J., Singer, Y. (eds) Learning Theory. COLT 2004. Lecture Notes in Computer Science(), vol 3120. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-540-27819-1_19
Download citation
DOI: https://doi.org/10.1007/978-3-540-27819-1_19
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-22282-8
Online ISBN: 978-3-540-27819-1
eBook Packages: Springer Book Archive