M. Malyutov

On Testing Degree of a Multivariate Polynomial Regression

This paper continues Separate Testing Inputs vs. Linear Programming relaxation. We study Response Surface Methodology in its decision stage on whether linear model should be replaced with a higher order polynomial model both under sparsity and its absence. Optimal properties of adding repeated measurements in the central point of the Complete Factorial Design under preliminary information on identical signs of quadratic coefficients is established under sparsity. Next, we study maximin designs which maximize the minimal power of discrimination under normality and a fixed norm of higher order coefficients.

КЛЮЧЕВЫЕ СЛОВА: Random design, active inputs, separate testing of inputs, sparsity, multivariate polynomial models, discrimination