Relaxed Clipping: A Global Training Method for Robust Regression and Classification

  title={Relaxed Clipping: A Global Training Method for Robust Regression and Classification},
  author={Yaoliang Yu and Min Yang and Linli Xu and Martha White and Dale Schuurmans},
Robust regression and classification are often thought to re quir non-convex loss functions that prevent scalable, global training. However , such a view neglects the possibility of reformulated training methods that can y ield practically solvable alternatives. A natural way to make a loss function more robu st to outliers is to truncate loss values that exceed a maximum threshold. We d emonstrate that a relaxation of this form of “loss clipping” can be made globa lly solvable and applicable… CONTINUE READING
Highly Cited
This paper has 34 citations. REVIEW CITATIONS
19 Citations
33 References
Similar Papers


Publications citing this paper.


Publications referenced by this paper.
Showing 1-10 of 33 references

Robust Statistics

  • P. Huber, E. Ronchetti
  • Wiley, 2nd edition
  • 2009
Highly Influential
6 Excerpts

Robust Statistics: Theory and Methods

  • R. Maronna, R. D. Martin, V. Yohai
  • Wiley
  • 2006
Highly Influential
8 Excerpts

Functional gradie nt techniques for combining hypotheses

  • L. Mason, J. Baxter, P. Bartlett, M. Frean
  • Advances in Large Margin Classifiers . MIT Press
  • 2000
Highly Influential
4 Excerpts

Robust Statistics: The Approach Based on Influence Functions

  • F. Hampel, E. Ronchetti, P. Rousseeuw, W. Stahel
  • Wiley
  • 1986
Highly Influential
4 Excerpts


  • J. Yu, S. Vishwanathan
  • G ünter, and N. Schraudolph. A quasi-Newton…
  • 2010
1 Excerpt

Similar Papers

Loading similar papers…