A Novel Nonparametric Maximum Likelihood Estimator for Probability Density Functions.

Area of Science:

• Statistics
• Machine Learning
• Data Science

Background:

• Parametric maximum likelihood (ML) estimators are efficient but struggle with complex data requiring nonparametric approaches.
• Existing nonparametric methods like kernel density estimation (KDE) lack desirable ML properties and have slower convergence rates.

Purpose of the Study:

• Introduce a new nonparametric ML estimator for probability density functions (pdfs).
• Address limitations of current nonparametric methods by leveraging band-limited assumptions for improved performance.

Main Methods:

• Developed a nonparametric ML estimator assuming the square-root of the pdf is band-limited (BLML).
• Computed the BLML estimator and demonstrated its consistency.
• Presented algorithms for efficient computation with lower complexity than KDE.

Main Results:

• The BLML estimator is shown to be consistent.
• Simulations indicate faster convergence rates compared to state-of-the-art nonparametric methods.
• The BLML estimator outperforms existing methods in density tail estimation and neuronal receptive field analysis.

Conclusions:

• The proposed BLML estimator offers a powerful new tool for nonparametric density estimation.
• It provides a consistent and efficient alternative to existing methods, particularly for complex datasets.
• Demonstrated practical utility in specialized applications within neuroscience and statistical analysis.