Sample-Adaptivity Tradeoff in On-Demand Sampling
Haghtalab, Nika, Montasser, Omar, Qiao, Mingda
We study the tradeoff between sample complexity and round complexity in on-demand sampling, where the learning algorithm adaptively samples from $k$ distributions over a limited number of rounds. In the realizable setting of Multi-Distribution Learning (MDL), we show that the optimal sample complexity of an $r$-round algorithm scales approximately as $dk^{Θ(1/r)} / ε$. For the general agnostic case, we present an algorithm that achieves near-optimal sample complexity of $\widetilde O((d + k) / ε^2)$ within $\widetilde O(\sqrt{k})$ rounds. Of independent interest, we introduce a new framework, Optimization via On-Demand Sampling (OODS), which abstracts the sample-adaptivity tradeoff and captures most existing MDL algorithms. We establish nearly tight bounds on the round complexity in the OODS setting. The upper bounds directly yield the $\widetilde O(\sqrt{k})$-round algorithm for agnostic MDL, while the lower bounds imply that achieving sub-polynomial round complexity would require fundamentally new techniques that bypass the inherent hardness of OODS.
Nov-20-2025
- Country:
- Africa > Ethiopia
- Addis Ababa > Addis Ababa (0.04)
- Asia
- Afghanistan > Parwan Province
- Charikar (0.04)
- India > Karnataka
- Bengaluru (0.04)
- Middle East > Jordan (0.04)
- Afghanistan > Parwan Province
- Europe
- Austria > Vienna (0.14)
- Netherlands > North Holland
- Amsterdam (0.04)
- Spain > Andalusia
- Cádiz Province > Cadiz (0.04)
- Sweden > Stockholm
- Stockholm (0.04)
- North America
- Canada
- United States
- Arizona > Maricopa County
- Phoenix (0.04)
- California
- Alameda County > Berkeley (0.04)
- Los Angeles County
- Long Beach (0.14)
- Los Angeles (0.14)
- San Diego County > San Diego (0.04)
- Colorado > Denver County
- Denver (0.04)
- Illinois > Cook County
- Chicago (0.04)
- Louisiana > Orleans Parish
- New Orleans (0.04)
- Maryland > Baltimore (0.04)
- Massachusetts > Hampshire County
- Amherst (0.04)
- Arizona > Maricopa County
- Africa > Ethiopia
- Genre:
- Research Report (0.81)
- Technology: