ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2406.06509
28
0

Robust Distribution Learning with Local and Global Adversarial Corruptions

10 June 2024
Sloan Nietert
Ziv Goldfeld
Soroosh Shafiee
    OOD
ArXivPDFHTML
Abstract

We consider learning in an adversarial environment, where an ε\varepsilonε-fraction of samples from a distribution PPP are arbitrarily modified (global corruptions) and the remaining perturbations have average magnitude bounded by ρ\rhoρ (local corruptions). Given access to nnn such corrupted samples, we seek a computationally efficient estimator P^n\hat{P}_nP^n​ that minimizes the Wasserstein distance W1(P^n,P)\mathsf{W}_1(\hat{P}_n,P)W1​(P^n​,P). In fact, we attack the fine-grained task of minimizing W1(Π#P^n,Π#P)\mathsf{W}_1(\Pi_\# \hat{P}_n, \Pi_\# P)W1​(Π#​P^n​,Π#​P) for all orthogonal projections Π∈Rd×d\Pi \in \mathbb{R}^{d \times d}Π∈Rd×d, with performance scaling with rank(Π)=k\mathrm{rank}(\Pi) = krank(Π)=k. This allows us to account simultaneously for mean estimation (k=1k=1k=1), distribution estimation (k=dk=dk=d), as well as the settings interpolating between these two extremes. We characterize the optimal population-limit risk for this task and then develop an efficient finite-sample algorithm with error bounded by εk+ρ+O~(dkn−1/(k∨2))\sqrt{\varepsilon k} + \rho + \tilde{O}(d\sqrt{k}n^{-1/(k \lor 2)})εk​+ρ+O~(dk​n−1/(k∨2)) when PPP has bounded covariance. This guarantee holds uniformly in kkk and is minimax optimal up to the sub-optimality of the plug-in estimator when ρ=ε=0\rho = \varepsilon = 0ρ=ε=0. Our efficient procedure relies on a novel trace norm approximation of an ideal yet intractable 2-Wasserstein projection estimator. We apply this algorithm to robust stochastic optimization, and, in the process, uncover a new method for overcoming the curse of dimensionality in Wasserstein distributionally robust optimization.

View on arXiv
Comments on this paper