ProbXiv
Explorersign in
Problem archiveProblem record

Statement

Let n≥2n\ge 2 independent arms have distributions Xi∼N(μi,1)X_i\sim\mathcal{N}(\mu_i,1), with μi∈[0,1]\mu_i\in[0,1] and a unique optimal arm ∗*. Write Δi=μ∗−μi\Delta_i=\mu_*-\mu_i for i≠∗i\ne *, H=∑i≠∗Δi−2H=\sum_{i\ne *}\Delta_i^{-2}, and, for each integer r≥0r\ge 0, let pr=H−1∑i≠∗: 2−(r+1)<Δi≤2−rΔi−2p_r=H^{-1}\sum_{i\ne *:\,2^{-(r+1)}<\Delta_i\le 2^{-r}}\Delta_i^{-2}. Define Ent⁡(I)=∑r:pr>0prlog⁡(1/pr)\operatorname{Ent}(I)=\sum_{r:p_r>0}p_r\log(1/p_r). The gap-entropy conjecture asserts that the optimal expected sample complexity of any algorithm that identifies ∗* with probability at least 1−δ1-\delta on every such instance is, up to universal constant factors and the unavoidable closest-gap term, H(log⁡(1/δ)+Ent⁡(I))H\bigl(\log(1/\delta)+\operatorname{Ent}(I)\bigr); equivalently, the instance-dependent lower and upper bounds should match at this order.

Record

Added
Links

Comments

Nothing has been published against this problem yet, and nothing has been said about it.

Sign in with an institutional address to take part in the discussion. Reading every thread stays open to everyone.

Sign in

Solve with an agent

Open the statement in a chat, with the problem and the ground rules already written into the prompt.

This opens a third-party site. Nothing is posted back to ProbXiv and nothing you write there is recorded here — what a model gives you is an attempt, which a person still has to check.