Parametric Classification of Bingham Distributions Based on Grassmann Manifolds

Area of Science:

• Computational Statistics
• Machine Learning
• Data Science

Background:

• Matrix variate Bingham distributions are complex statistical models.
• Accurate classification requires robust estimation of distribution parameters.
• Existing methods for Bingham distributions have limitations in parameter estimation and computational efficiency.

Purpose of the Study:

• To develop a novel Bayesian classification framework for matrix variate Bingham distributions.
• To introduce a consistent parametric modeling framework utilizing Grassmann manifolds.
• To enhance the calculation of normalizing constants for Bingham models.

Main Methods:

• Extended saddle-point approximation (SPA) for calculating normalizing constants.
• Employed maximum likelihood estimation (MLE) for parameter evaluation.
• Developed a general parametric modeling framework based on Grassmann manifolds.

Main Results:

• The proposed Bayesian classification framework shows superior performance on most of the 14 tested real-world visual classification databases.
• The extended SPA method effectively calculates normalizing constants for Bingham distributions.
• The MLE approach accurately estimates parameters within the probability density functions.

Conclusions:

• The novel Bayesian classification framework offers improved accuracy and performance for matrix variate Bingham distributions.
• The integration of SPA and MLE provides a powerful tool for statistical modeling and classification.
• This research advances the field of directional statistics and machine learning applications.