Local linear smoothing for regression surfaces on the simplex using Dirichlet kernels

2024-08-13

AI Summary

This paper addresses the challenges of nonparametric regression surface estimation on the simplex domain, particularly severe boundary bias and incompatible weighting in conventional methods. We propose a novel local linear smoother employing a locally adaptive Dirichlet kernel. To our knowledge, this is the first work to integrate local linear fitting with the Dirichlet kernel, enabling rigorous asymptotic derivation of the estimator’s bias, variance, and mean integrated squared error (MISE), thereby achieving a theoretical breakthrough in boundary correction. The method inherently respects the simplex geometry—requiring no boundary reflection or domain transformation. Monte Carlo simulations demonstrate that it reduces mean squared error at boundary points by over 30% compared to the classical Dirichlet-kernel Nadaraya–Watson estimator. Furthermore, the theoretical framework extends naturally to higher-dimensional simplices. Our approach establishes a new paradigm for simplex-constrained nonparametric regression, balancing accuracy, robustness, and interpretability.

Technology Category

• Machine Learning: Kernel Methods
• Search and Optimization: Non-convex Optimization
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

This paper introduces a local linear smoother for regression surfaces on the simplex. The estimator solves a least-squares regression problem weighted by a locally adaptive Dirichlet kernel, ensuring good boundary properties. Asymptotic results for the bias, variance, mean squared error, and mean integrated squared error are derived, generalizing the univariate results of Chen [Ann. Inst. Statist. Math., 54(2) (2002), pp. 312-323]. A simulation study shows that the proposed local linear estimator with Dirichlet kernel outperforms its only direct competitor in the literature, the Nadaraya-Watson estimator with Dirichlet kernel due to Bouzebda, Nezzal and Elhattab [AIMS Math., 9(9) (2024), pp. 26195-26282].

Problem

• Develops local linear smoother for simplex regression surfaces
• Uses adaptive Dirichlet kernel for boundary properties
• Outperforms Nadaraya-Watson estimator in simulations

Innovation

• Local linear smoother for simplex regression surfaces
• Dirichlet kernel ensures good boundary properties
• Outperforms Nadaraya-Watson estimator in simulations