2016

Does Distributionally Robust Supervised Learning Give Robust Classifiers?

Hu, Weihua, Niu, Gang, Sato, Issei et al.

Understand

Distributionally Robust Supervised Learning (DRSL) is necessary for building reliable machine learning systems.

When machine learning is deployed in the real world, its performance can be significantly degraded because test data may follow a different distribution from training data.
DRSL with f-divergences explicitly considers the worst-case distribution shift by minimizing the adversarially reweighted training loss.
In this paper, we analyze this DRSL, focusing on the classification scenario.

The theory of max-min, with applications
Danskin, J. M · 1966
Earlier work this paper cites.
Information-type measures of difference of probability distributions and indirect observations
Ciszar, I · 1967
Earlier work this paper cites.
A Course in Probability Theory
Chung, K. L · 1968
Earlier work this paper cites.
A characterization of stability in linear programming
Robinson, S. M · 1977
Earlier work this paper cites.
Newsweeder: Learning to filter netnews
Lang, K · 1995
Earlier work this paper cites.
UCI repository of machine learning databases
Blake, C. and Merz, C. J · 1998
Earlier work this paper cites.
Optimization problems with perturbations, a guided tour
Bonnans, J. F. and Shapiro, A · 1998
Earlier work this paper cites.
Gradient-based learning applied to document recognition
LeCun, Y., Bottou, L., Bengio, Y., and Haffner, P · 1998
Earlier work this paper cites.
Statistical Learning Theory
Vapnik, V. N · 1998
Earlier work this paper cites.

Adjusting the outputs of a classifier to new a priori probabilities: a simple procedure
Saerens, M., Latinne, P., and Decaestecker, C · 2002
Cited alongside, same era.
Robust supervised learning
Bagnell, J. A · 2005
Cited alongside, same era.
Convexity, classification, and risk bounds
Bartlett, P. L., Jordan, M. I., and McAuliffe, J. D · 2006
Cited alongside, same era.
Nightmare at test time: robust learning by feature deletion
Globerson, A. and Roweis, S · 2006
Cited alongside, same era.
Mixture regression for covariate shift
Storkey, A. J. and Sugiyama, M · 2007
Cited alongside, same era.
On the consistency of multiclass classification methods
Tewari, A. and Bartlett, P. L · 2007
Cited alongside, same era.
Dataset shift in machine learning
Quionero-Candela, J., Sugiyama, M., Schwaighofer, A., and Lawrence, N. D · 2009
Cited alongside, same era.
Robust solutions of optimization problems affected by uncertain probabilities
Ben-Tal, A., Den Hertog, D., De Waegenaere, A., Melenberg, B., and Rennen, G · 2013
Cited alongside, same era.
Robust classification under sample selection bias
Liu, A. and Ziebart, B · 2014
Cited alongside, same era.

Data-driven distributionally robust optimization using the wasserstein metric: Performance guarantees and tractable reformulations
Original
Esfahani, P. M. and Kuhn, D · 2015
Later among the works it cites.
From categories to subcategories: large-scale image classification with partial class label refinement
Ristin, M., Gall, J., Guillaumin, M., and Van Gool, L · 2015
Later among the works it cites.
Robust wasserstein profile inference and applications to machine learning
Original
Blanchet, J., Kang, Y., and Murthy, K · 2016
Closest in time.
Statistics of robust optimization: A generalized empirical likelihood approach
Original
Duchi, J., Glynn, P., and Namkoong, H · 2016
Closest in time.
Stochastic gradient methods for distributionally robust optimization with f-divergences
Namkoong, H. and Duchi, J. C · 2016
Closest in time.
Variance-based regularization with convex objectives
Namkoong, H. and Duchi, J. C · 2017
Closest in time.
Certifiable distributional robustness with principled adversarial training
Original
Sinha, A., Namkoong, H., and Duchi, J · 2017
Closest in time.
Fairness without demographics in repeated loss minimization
Hashimoto, T. B., Srivastava, M., Namkoong, H., and Liang, P · 2018
Closest in time.

alphaXiv searches the wider corpus for related work and actual follow-ups.

alphaXiv is searching for related work…

Reading the bibliography…