Abstract
We study the problem of learning partitions using equivalence constraints as input. This is a binary classification problem in the product space of pairs of datapoints. The training data includes pairs of datapoints which are labeled as coming from the same class or not. This kind of data appears naturally in applications where explicit labeling of datapoints is hard to get, but relations between datapoints can be more easily obtained, using, for example, Markovian dependency (as in video clips).
Our problem is an unlabeled partition problem, and is therefore tightly related to multiclass classification. We show that the solutions of the two problems are related, in the sense that a good solution to the binary classification problem entails the existence of a good solution to the multiclass problem, and vice versa. We also show that bounds on the sample complexity of the two problems are similar, by showing that their relevant ‘dimensions’ (VC dimension for the binary problem, Natarajan dimension for the multiclass problem) bound each other. Finally, we show the feasibility of solving multiclass learning efficiently by using a solution of the equivalent binary classification problem. In this way advanced techniques developed for binary classification, such as SVM and boosting, can be used directly to enhance multiclass learning.
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
Allwein, E., Schapire, R., Singer, Y.: Reducing Multiclass to Bin ary: A Unifying Approach for Margin Classifiers. Journal of Machine Learning Research 1, 113–141 (2000)
Bansal, N., Blum, A., Chawla, S.: Correlation Clustering. In: Proc. FOCS, pp. 238-247 (2002)
Bar-Hillel, A., Weinshall, D.: Learning with Equivalence Constraints. HU Technical Report 2003-38, in http://www.cs.huji.ac.il/~daphna
Ben-David, S., Cesa-Bianchi, N., Haussler, D., Long, P.H.: Characterizations of learnability for classes of 0,......., n-valued functions
Dietterich, T.G., Bakiri, G.: Solving multiclass learning problems via errorcorrecting output codes. Journal of Artifical Intelligence Research 2, 263–286 (1995)
Guruswami, V., Sahai, A.: Multiclass learning, Boosting, and Error- Correcting codes. In: Proc. COLT (1999)
Har-Peled, S., Roth, D., Zimak, D.: Constraints classification:A new approach to multiclass classification and ranking. In: Proc. NIPS (2002)
Haussler, D., Long, P.M.: A generalization of Sauer’s lemma. Technical Report UCSU-CRL-90-15. UC Santa Cruz (1990)
Kearns, M.J., Vazirani, U.V.: An Introduction to Computational Learning Theory. MIT Press, Cambridge (1994)
Natarajan, B.K.: On learning sets and functions. Machine Learning 4, 67–97 (1989)
Phillips, P.J.: Support vector machines applied to face recognition. In: Mozer, M.C., Jordan, M.I., Petsche, T. (eds.) Advances in Neural Information Processing Systems, vol. 11, p. 803. MIT Press, Cambridge (1998)
Hertz, T., Shental, N., Bar-Hillel, A., Weinshall, D.: Enhancing Image and Video Retrieval: Learning via Equivalence Constraints. In: Proc. of IEEE Conference on Computer Vision and Pattern Recognition (2003)
Schapire, R.E.: A brief introduction to boosting. In: Proc. of the 16th International Joint Conference on Artificial Intelligence (1999)
Vapnik, V.N.: The Nature of Statistical Learning. Springer, Heidelberg (1995)
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
Bar-Hillel, A., Weinshall, D. (2003). Learning with Equivalence Constraints and the Relation to Multiclass Learning. 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_46
Download citation
DOI: https://doi.org/10.1007/978-3-540-45167-9_46
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-40720-1
Online ISBN: 978-3-540-45167-9
eBook Packages: Springer Book Archive