On Learning Distributions from their Samples

Sudeep Kamath, Alon Orlitsky, Dheeraj Pichapati, Ananda Theertha Suresh
Proceedings of The 28th Conference on Learning Theory, PMLR 40:1066-1100, 2015.

Abstract

One of the most natural and important questions in statistical learning is: how well can a distribution be approximated from its samples. Surprisingly, this question has so far been resolved for only one loss, the KL-divergence and even in this case, the estimator used is ad hoc and not well understood. We study distribution approximations for general loss measures. For \ell_2^2 we determine the best approximation possible, for \ell_1 and χ^2 we derive tight bounds on the best approximation, and when the probabilities are bounded away from zero, we resolve the question for all sufficiently smooth loss measures, thereby providing a coherent understanding of the rate at which distributions can be approximated from their samples.

Cite this Paper


BibTeX
@InProceedings{pmlr-v40-Kamath15, title = {On Learning Distributions from their Samples}, author = {Kamath, Sudeep and Orlitsky, Alon and Pichapati, Dheeraj and Suresh, Ananda Theertha}, booktitle = {Proceedings of The 28th Conference on Learning Theory}, pages = {1066--1100}, year = {2015}, editor = {Grünwald, Peter and Hazan, Elad and Kale, Satyen}, volume = {40}, series = {Proceedings of Machine Learning Research}, address = {Paris, France}, month = {03--06 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v40/Kamath15.pdf}, url = {https://proceedings.mlr.press/v40/Kamath15.html}, abstract = {One of the most natural and important questions in statistical learning is: how well can a distribution be approximated from its samples. Surprisingly, this question has so far been resolved for only one loss, the KL-divergence and even in this case, the estimator used is ad hoc and not well understood. We study distribution approximations for general loss measures. For \ell_2^2 we determine the best approximation possible, for \ell_1 and χ^2 we derive tight bounds on the best approximation, and when the probabilities are bounded away from zero, we resolve the question for all sufficiently smooth loss measures, thereby providing a coherent understanding of the rate at which distributions can be approximated from their samples.} }
Endnote
%0 Conference Paper %T On Learning Distributions from their Samples %A Sudeep Kamath %A Alon Orlitsky %A Dheeraj Pichapati %A Ananda Theertha Suresh %B Proceedings of The 28th Conference on Learning Theory %C Proceedings of Machine Learning Research %D 2015 %E Peter Grünwald %E Elad Hazan %E Satyen Kale %F pmlr-v40-Kamath15 %I PMLR %P 1066--1100 %U https://proceedings.mlr.press/v40/Kamath15.html %V 40 %X One of the most natural and important questions in statistical learning is: how well can a distribution be approximated from its samples. Surprisingly, this question has so far been resolved for only one loss, the KL-divergence and even in this case, the estimator used is ad hoc and not well understood. We study distribution approximations for general loss measures. For \ell_2^2 we determine the best approximation possible, for \ell_1 and χ^2 we derive tight bounds on the best approximation, and when the probabilities are bounded away from zero, we resolve the question for all sufficiently smooth loss measures, thereby providing a coherent understanding of the rate at which distributions can be approximated from their samples.
RIS
TY - CPAPER TI - On Learning Distributions from their Samples AU - Sudeep Kamath AU - Alon Orlitsky AU - Dheeraj Pichapati AU - Ananda Theertha Suresh BT - Proceedings of The 28th Conference on Learning Theory DA - 2015/06/26 ED - Peter Grünwald ED - Elad Hazan ED - Satyen Kale ID - pmlr-v40-Kamath15 PB - PMLR DP - Proceedings of Machine Learning Research VL - 40 SP - 1066 EP - 1100 L1 - http://proceedings.mlr.press/v40/Kamath15.pdf UR - https://proceedings.mlr.press/v40/Kamath15.html AB - One of the most natural and important questions in statistical learning is: how well can a distribution be approximated from its samples. Surprisingly, this question has so far been resolved for only one loss, the KL-divergence and even in this case, the estimator used is ad hoc and not well understood. We study distribution approximations for general loss measures. For \ell_2^2 we determine the best approximation possible, for \ell_1 and χ^2 we derive tight bounds on the best approximation, and when the probabilities are bounded away from zero, we resolve the question for all sufficiently smooth loss measures, thereby providing a coherent understanding of the rate at which distributions can be approximated from their samples. ER -
APA
Kamath, S., Orlitsky, A., Pichapati, D. & Suresh, A.T.. (2015). On Learning Distributions from their Samples. Proceedings of The 28th Conference on Learning Theory, in Proceedings of Machine Learning Research 40:1066-1100 Available from https://proceedings.mlr.press/v40/Kamath15.html.

Related Material