Estimation 2 - The MAP Estimator – Regularization & Point Estimates

This video provides a deep dive into the Maximum A Posteriori (MAP) estimator, a point estimator which finds the specific parameter value that maximizes the posterior distribution. We explain the mathematical formula, showing how you calculate the argmax of the posterior by taking the derivative of the likelihood times the prior. We also explore the close relationship between MAP and Maximum Likelihood (ML), demonstrating that if you use a uniform or "flat" non-informative prior, the MAP estimate is exactly equal to the ML estimate. Furthermore, we explain how incorporating an informative prior pulls the estimate and acts as a built-in regularization for the ML method. Finally, we contrast MAP's single point estimate with the more powerful—but computationally expensive—full Bayesian approach, which utilizes the entire posterior distribution (often via Markov Chain Monte Carlo sampling) to calculate Bayesian confidence intervals, known as credible intervals.

Timestamps / Chapters:
00:00 - The MAP Estimator & Formula: Finding the mode of the posterior distribution using the argmax function.
01:30 - Relationship to Maximum Likelihood: Why MAP equals ML when using a uniform, non-informative prior.
03:00 - MAP as Regularization: How incorporating a prior pulls the estimate and acts as a regularization technique for ML estimation.
04:30 - Point Estimates vs. Full Distributions: Understanding the difference between the point estimates of MAP and ML compared to exploring the full posterior distribution.
06:00 - Credible Intervals & Computational Cost: Extracting credible intervals from the full posterior, and why using methods like MCMC is computationally demanding compared to simple point estimation

Видео Estimation 2 - The MAP Estimator – Regularization & Point Estimates канала Khoka