Polynomial-time attainment of the noiseless minimax rate
Determine whether any polynomial-time simplex estimator can attain the noiseless minimax sample-complexity rate \(\Theta(K/N)\) under the uniform simplex-sampling model.
References
In this paper we propose Deep Minimum Volume Simplex Analysis (DeepMVSA), a neural implicit simplex estimator whose memory footprint and parameter complexity are independent of $N$, and we answer all three questions for it; whether any polynomial-time estimator can attain the noiseless minimax rate $\Theta(K/N)$ remains open.
The experiments therefore probe the mechanisms identified by the theory---the approximation floor of Remark~\ref{rem:apxfloor}, the capacity prescription of Proposition~\ref{prop:apxrate}, and the scaling laws of Theorem~\ref{thm:complexity}---rather than verifying the theorem's constants; narrowing the gap between the combinatorial threshold, inherited from the window argument of Lemma~S2 of , and practical sample sizes remains open.