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. 2401.15520
15
0

Oracle-Efficient Hybrid Online Learning with Unknown Distribution

27 January 2024
Changlong Wu
Jin Sima
Wojtek Szpankowski
ArXivPDFHTML
Abstract

We study the problem of oracle-efficient hybrid online learning when the features are generated by an unknown i.i.d. process and the labels are generated adversarially. Assuming access to an (offline) ERM oracle, we show that there exists a computationally efficient online predictor that achieves a regret upper bounded by O~(T34)\tilde{O}(T^{\frac{3}{4}})O~(T43​) for a finite-VC class, and upper bounded by O~(Tp+1p+2)\tilde{O}(T^{\frac{p+1}{p+2}})O~(Tp+2p+1​) for a class with α\alphaα fat-shattering dimension α−p\alpha^{-p}α−p. This provides the first known oracle-efficient sublinear regret bounds for hybrid online learning with an unknown feature generation process. In particular, it confirms a conjecture of Lazaric and Munos (JCSS 2012). We then extend our result to the scenario of shifting distributions with KKK changes, yielding a regret of order O~(T45K15)\tilde{O}(T^{\frac{4}{5}}K^{\frac{1}{5}})O~(T54​K51​). Finally, we establish a regret of O~((K23(log⁡∣H∣)13+K)⋅T45)\tilde{O}((K^{\frac{2}{3}}(\log|\mathcal{H}|)^{\frac{1}{3}}+K)\cdot T^{\frac{4}{5}})O~((K32​(log∣H∣)31​+K)⋅T54​) for the contextual KKK-armed bandits with a finite policy set H\mathcal{H}H, i.i.d. generated contexts from an unknown distribution, and adversarially generated costs.

View on arXiv
Comments on this paper