This paper introduces a sharp global dynamic analogue to Haussler's covering lemma, extending it to bounded real-valued classes under normalized $\ell_p$ distance for $1 \le p < \infty$. The research provides a global dynamic cover bound that is independent of the instance sequence and offers matching worst-case lower bounds for binary classes. The proof method combines rank-decreasing correction schedules with accurate sampled prefixes and exact continuations, utilizing shared-parent routing to optimize these continuations. AI
RANK_REASON This is a research paper published on arXiv detailing theoretical advancements in machine learning. [lever_c_demoted from research: ic=1 ai=1.0]
- ACM Symposium on Theory of Computing
- Alon
- CAVIN1
- Dagan
- Haussler's covering lemma
- JcTaxWT-Neo7
- Matthias: Das Berliner Schloss
- Michael/WEMHOFF
- Moran
- Naor
- Rakhlin
- Sridharan
- Tewari
- Tobiah ben Eliezer
- Trans. AMS
- VC classes
- Yogev
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →