Approximation by multivariate Bernstein-Durrmeyer operators and learning rates of least-squares regularized regression with multivariate polynomial kernels (Q390534): Difference between revisions

Approximations by Bernstein-Durrmayer operators with respect to Lebesque measures on a simplex \(X\subset \mathbb{R}^n\) have been studied for a long time. On the whole, the approximation properties with respect to Jacobi weights are studied. The problem of uniform convergence with respect to a strictly positive Borel probability measure \(\rho _X\) was already studied by \textit{E. E. Berdysheva} [J. Math. Anal. Appl. 394, No. 1, 324--336 (2012; Zbl 1247.41014)]. Here, the approximation problems by Bernstein-Durrmayer operators in \(L^p_{\rho _X}\), \(1\leq p<\infty \), with an arbitrary Borel probability measure are studied. Convergence rates of type \(O(m^{-\gamma })\) with some \(\gamma >0\) for the least-squares regularized regression algorithm associated with a multivariate polynomial kernel (where \(m\) is the simple size) are obtained. The learning rates depend on the space dimension \(n\) and the capacity of the reproducing kernel Hilbert space generated by the polynomial kernel.