-
An improved nonparametric test and sample size procedures for the randomized complete block designs
Authors:
Show-Li Jan,
Gwowen Shieh
Abstract:
The Friedman test has been extensively applied as a nonparametric alternative to the conventional F procedure for comparing treatment effects in randomized complete block designs. A chi-square distribution provides a convenient approximation to determining the critical values for the Friedman procedure in hypothesis testing. However, the chi-square approximation is generally conservative and the a…
▽ More
The Friedman test has been extensively applied as a nonparametric alternative to the conventional F procedure for comparing treatment effects in randomized complete block designs. A chi-square distribution provides a convenient approximation to determining the critical values for the Friedman procedure in hypothesis testing. However, the chi-square approximation is generally conservative and the accuracy declines with increasing number of treatments. This paper describes an alternative transformation of the Friedman statistic along with an approximate F distribution that has the same numerator degrees of freedom as the ANOVA F test. Moreover, two approximate noncentral F distributions are presented for the proposed F-transformation under the alternative hypothesis of heterogeneous location shifts. Explicit power functions are derived when the underlying populations have the uniform, normal, Laplace, and exponential distributions. Theoretical examination and empirical assessment are presented to validate the advantages of the proposed approaches over the existing methods of the Friedman test. The developed test and power procedures are recommended due to their consistently acceptable Type I error rates and accurate power calculations for the location shift structures and population distributions considered here.
△ Less
Submitted 21 March, 2025;
originally announced March 2025.
-
MIQCQP reformulation of the ReLU neural networks Lipschitz constant estimation problem
Authors:
Mohammed Sbihi,
Sophie Jan,
Nicolas Couellan
Abstract:
It is well established that to ensure or certify the robustness of a neural network, its Lipschitz constant plays a prominent role. However, its calculation is NP-hard. In this note, by taking into account activation regions at each layer as new constraints, we propose new quadratically constrained MIP formulations for the neural network Lipschitz estimation problem. The solutions of these problem…
▽ More
It is well established that to ensure or certify the robustness of a neural network, its Lipschitz constant plays a prominent role. However, its calculation is NP-hard. In this note, by taking into account activation regions at each layer as new constraints, we propose new quadratically constrained MIP formulations for the neural network Lipschitz estimation problem. The solutions of these problems give lower bounds and upper bounds of the Lipschitz constant and we detail conditions when they coincide with the exact Lipschitz constant.
△ Less
Submitted 2 February, 2024;
originally announced February 2024.
-
The layer number of grids
Authors:
Gergely Ambrus,
Alexander Hsu,
Bo Peng,
Shiyu Jan
Abstract:
The peeling process is defined as follows: starting with a finite point set $X \subset \mathbb{R}^d$, we repeatedly remove the set of vertices of the convex hull of the current set of points. The number of peeling steps needed to completely delete the set $X$ is called the layer number of $X$. In this paper, we study the layer number of the $d$-dimensional integer grid $[n]^d$. We prove that for e…
▽ More
The peeling process is defined as follows: starting with a finite point set $X \subset \mathbb{R}^d$, we repeatedly remove the set of vertices of the convex hull of the current set of points. The number of peeling steps needed to completely delete the set $X$ is called the layer number of $X$. In this paper, we study the layer number of the $d$-dimensional integer grid $[n]^d$. We prove that for every $d \geq 1$, the layer number of $[n]^d$ is at least $Ω\left(n^\frac{2d}{d+1}\right)$. On the other hand, we show that for every $d\geq 3$, it takes at most $O(n^{d - 9/11})$ steps to fully remove $[n]^d$. Our approach is based on an enhancement of the method used by Har-Peled and Lidický for solving the 2-dimensional case.
△ Less
Submitted 28 September, 2020;
originally announced September 2020.
