Learning Models with Uniform Performance via Distributionally Robust Optimization
This addresses the challenge of model robustness to distribution shifts for applications in statistics and machine learning, though it appears incremental as it builds on existing DRO methods.
The paper tackles the problem of learning models that perform well under distributional shifts by developing a distributionally robust optimization (DRO) framework, resulting in improved performance on real tasks like generalizing to unknown subpopulations and fine-grained recognition.
A common goal in statistics and machine learning is to learn models that can perform well against distributional shifts, such as latent heterogeneous subpopulations, unknown covariate shifts, or unmodeled temporal effects. We develop and analyze a distributionally robust stochastic optimization (DRO) framework that learns a model providing good performance against perturbations to the data-generating distribution. We give a convex formulation for the problem, providing several convergence guarantees. We prove finite-sample minimax upper and lower bounds, showing that distributional robustness sometimes comes at a cost in convergence rates. We give limit theorems for the learned parameters, where we fully specify the limiting distribution so that confidence intervals can be computed. On real tasks including generalizing to unknown subpopulations, fine-grained recognition, and providing good tail performance, the distributionally robust approach often exhibits improved performance.