RobustOpt
Code for solving robust stochastic---or distributionally robust---optimization problems with f-divergences.
Efficient computation of full-batch gradient for the robust loss
The files simple_projections.py, SimpleProjections.jl each contain efficient binary search algorithm for solving the inner worst-case problem of the robust formulation. See Appendix C of this paper for a full treatment of the algorithm.
Papers
- Variance Regularization with Convex Objectives (NIPS Version, Long Version)
- Statistics of Robust Optimization: A Generalized Empirical Likelihood Approach (arXiv link)
See also Stochastic Gradient Methods for Distributionally Robust Optimization with f-divergences (link) for a stochastic gradient procedure.
Authors
License
This project is licensed under the MIT License - see the LICENSE.md file for details