← All problems

Conjecture on the block max-min estimator's pivotal limit distribution

6 years old · traced to

Let f0f_0 be the monotone regression function, let x0x_0 be the point of inference, and let f^n−(x0)\widehat{f}_n^{-}(x_0) be the block max-min estimator with associated block count nu^,v^−(x0)n_{\widehat{u},\widehat{v}}^{-}(x_0). Let σ\sigma denote the noise standard deviation, and let α\bm{\alpha} and K(f0,x0)K(f_0,x_0) be as in Theorem. Under the same settings as that theorem, there is a finite random variable Lα−\mathbb{L}_{\bm{\alpha}}^{-} that does not depend on K(f0,x0)K(f_0,x_0). Block max-min pivotal-limit conjecture.

nu^,v^−(x0)(f^n−(x0)−f0(x0))⇝σ⋅Lα−.\sqrt{n_{\widehat{u},\widehat{v}}^{-}(x_0)}\big(\widehat{f}_n^{-}(x_0)-f_0(x_0)\big)\rightsquigarrow \sigma\cdot \mathbb{L}_{\bm{\alpha}}^{-}.

If true, this would provide an asymptotic pivotal quantity for confidence intervals based on the block max-min estimator alone, potentially reducing the computational cost relative to the block average procedure. The source gives no resolution of this conjecture.

References
  1. Hang Deng, Qiyang Han and Cun-Hui Zhang, “Confidence intervals for multiple isotonic regression and other monotone models”, arXiv:2001.07064 (2020).

Progress summary

Never refreshed

Nothing recorded yet. Refresh searches the literature and the public web for attempts on this problem, and writes the first summary here.

Solutions 0

No solutions have been posted yet.