Learning with Risks Based on M-Location

Abstract

In this work, we study a new class of risks defined in terms of the location and deviation of the loss distribution, generalizing far beyond classical mean-variance risk functions. The class is easily implemented as a wrapper around any smooth loss, it admits finite-sample stationarity guarantees for stochastic gradient methods, it is straightforward to interpret and adjust, with close links to M-estimators of the loss location, and has a salient effect on the test loss distribution, giving us control over symmetry and deviations that are not possible under naive ERM.

Cite

Text

Holland. "Learning with Risks Based on M-Location." Machine Learning, 2022. doi:10.1007/S10994-022-06217-5

Markdown

[Holland. "Learning with Risks Based on M-Location." Machine Learning, 2022.](https://mlanthology.org/mlj/2022/holland2022mlj-learning/) doi:10.1007/S10994-022-06217-5

BibTeX

@article{holland2022mlj-learning,
  title     = {{Learning with Risks Based on M-Location}},
  author    = {Holland, Matthew J.},
  journal   = {Machine Learning},
  year      = {2022},
  pages     = {4679-4718},
  doi       = {10.1007/S10994-022-06217-5},
  volume    = {111},
  url       = {https://mlanthology.org/mlj/2022/holland2022mlj-learning/}
}