Heavy-tailed regression with a generalized median-of-means

Daniel Hsu, Sivan Sabato
; Proceedings of the 31st International Conference on Machine Learning, PMLR 32(2):37-45, 2014.

Abstract

This work proposes a simple and computationally efficient estimator for linear regression, and other smooth and strongly convex loss minimization problems. We prove loss approximation guarantees that hold for general distributions, including those with heavy tails. All prior results only hold for estimators which either assume bounded or subgaussian distributions, require prior knowledge of distributional properties, or are not known to be computationally tractable. In the special case of linear regression with possibly heavy-tailed responses and with bounded and well-conditioned covariates in d-dimensions, we show that a random sample of size \tildeO(d\log(1/δ)) suffices to obtain a constant factor approximation to the optimal loss with probability 1-δ, a minimax optimal sample complexity up to log factors. The core technique used in the proposed estimator is a new generalization of the median-of-means estimator to arbitrary metric spaces.

Cite this Paper


BibTeX
@InProceedings{pmlr-v32-hsu14, title = {Heavy-tailed regression with a generalized median-of-means}, author = {Daniel Hsu and Sivan Sabato}, booktitle = {Proceedings of the 31st International Conference on Machine Learning}, pages = {37--45}, year = {2014}, editor = {Eric P. Xing and Tony Jebara}, volume = {32}, number = {2}, series = {Proceedings of Machine Learning Research}, address = {Bejing, China}, month = {22--24 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v32/hsu14.pdf}, url = {http://proceedings.mlr.press/v32/hsu14.html}, abstract = {This work proposes a simple and computationally efficient estimator for linear regression, and other smooth and strongly convex loss minimization problems. We prove loss approximation guarantees that hold for general distributions, including those with heavy tails. All prior results only hold for estimators which either assume bounded or subgaussian distributions, require prior knowledge of distributional properties, or are not known to be computationally tractable. In the special case of linear regression with possibly heavy-tailed responses and with bounded and well-conditioned covariates in d-dimensions, we show that a random sample of size \tildeO(d\log(1/δ)) suffices to obtain a constant factor approximation to the optimal loss with probability 1-δ, a minimax optimal sample complexity up to log factors. The core technique used in the proposed estimator is a new generalization of the median-of-means estimator to arbitrary metric spaces.} }
Endnote
%0 Conference Paper %T Heavy-tailed regression with a generalized median-of-means %A Daniel Hsu %A Sivan Sabato %B Proceedings of the 31st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2014 %E Eric P. Xing %E Tony Jebara %F pmlr-v32-hsu14 %I PMLR %J Proceedings of Machine Learning Research %P 37--45 %U http://proceedings.mlr.press %V 32 %N 2 %W PMLR %X This work proposes a simple and computationally efficient estimator for linear regression, and other smooth and strongly convex loss minimization problems. We prove loss approximation guarantees that hold for general distributions, including those with heavy tails. All prior results only hold for estimators which either assume bounded or subgaussian distributions, require prior knowledge of distributional properties, or are not known to be computationally tractable. In the special case of linear regression with possibly heavy-tailed responses and with bounded and well-conditioned covariates in d-dimensions, we show that a random sample of size \tildeO(d\log(1/δ)) suffices to obtain a constant factor approximation to the optimal loss with probability 1-δ, a minimax optimal sample complexity up to log factors. The core technique used in the proposed estimator is a new generalization of the median-of-means estimator to arbitrary metric spaces.
RIS
TY - CPAPER TI - Heavy-tailed regression with a generalized median-of-means AU - Daniel Hsu AU - Sivan Sabato BT - Proceedings of the 31st International Conference on Machine Learning PY - 2014/01/27 DA - 2014/01/27 ED - Eric P. Xing ED - Tony Jebara ID - pmlr-v32-hsu14 PB - PMLR SP - 37 DP - PMLR EP - 45 L1 - http://proceedings.mlr.press/v32/hsu14.pdf UR - http://proceedings.mlr.press/v32/hsu14.html AB - This work proposes a simple and computationally efficient estimator for linear regression, and other smooth and strongly convex loss minimization problems. We prove loss approximation guarantees that hold for general distributions, including those with heavy tails. All prior results only hold for estimators which either assume bounded or subgaussian distributions, require prior knowledge of distributional properties, or are not known to be computationally tractable. In the special case of linear regression with possibly heavy-tailed responses and with bounded and well-conditioned covariates in d-dimensions, we show that a random sample of size \tildeO(d\log(1/δ)) suffices to obtain a constant factor approximation to the optimal loss with probability 1-δ, a minimax optimal sample complexity up to log factors. The core technique used in the proposed estimator is a new generalization of the median-of-means estimator to arbitrary metric spaces. ER -
APA
Hsu, D. & Sabato, S.. (2014). Heavy-tailed regression with a generalized median-of-means. Proceedings of the 31st International Conference on Machine Learning, in PMLR 32(2):37-45

Related Material