We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: On the power of adaptivity in statistical adversaries

Abstract: We study a fundamental question concerning adversarial noise models in statistical problems where the algorithm receives i.i.d. draws from a distribution $\mathcal{D}$. The definitions of these adversaries specify the type of allowable corruptions (noise model) as well as when these corruptions can be made (adaptivity); the latter differentiates between oblivious adversaries that can only corrupt the distribution $\mathcal{D}$ and adaptive adversaries that can have their corruptions depend on the specific sample $S$ that is drawn from $\mathcal{D}$.
In this work, we investigate whether oblivious adversaries are effectively equivalent to adaptive adversaries, across all noise models studied in the literature. Specifically, can the behavior of an algorithm $\mathcal{A}$ in the presence of oblivious adversaries always be well-approximated by that of an algorithm $\mathcal{A}'$ in the presence of adaptive adversaries? Our first result shows that this is indeed the case for the broad class of statistical query algorithms, under all reasonable noise models. We then show that in the specific case of additive noise, this equivalence holds for all algorithms. Finally, we map out an approach towards proving this statement in its fullest generality, for all algorithms and under all reasonable noise models.
Comments: To appear in COLT 2022. Updated with reviewer feedback
Subjects: Machine Learning (cs.LG); Computational Complexity (cs.CC); Data Structures and Algorithms (cs.DS)
Cite as: arXiv:2111.10352 [cs.LG]
  (or arXiv:2111.10352v2 [cs.LG] for this version)

Submission history

From: Guy Blanc [view email]
[v1] Fri, 19 Nov 2021 18:26:25 GMT (32kb)
[v2] Wed, 29 Jun 2022 14:03:34 GMT (57kb)

Link back to: arXiv, form interface, contact.