A0556
Title: Robust Bayesian estimation using divergences for von Mises-Fisher distribution
Authors: Tomoyuki Nakagawa - Meisei University (Japan) [presenting]
Sho Kazari - Tokyo University of Science (Japan)
Yasuhito Tsuruta - Meiji University (Japan)
Kouji Tahata - Tokyo University of Science (Japan)
Abstract: When outliers are present in the data, estimators such as the maximum likelihood estimator can exhibit substantial bias. This issue is no exception for directional data. In such cases, robust estimation methods have been proposed using Huber-type loss functions, generalized residuals, and divergence measures. However, it is often difficult to evaluate the asymptotic properties and uncertainty of these methods. Bayesian estimation is considered based on robust divergences, specifically the density power divergence and gamma-divergence. Sampling methods are proposed, the robustness of the estimators is examined, and their performance is evaluated through numerical experiments.
