VIII. Maximum likelihood and maximum a posteriori estimates
Given by Sven Laur
Brief summary: Formalisation of Maximum Likelihood and Maximum A Posteriori principles. Simple examples of ML and MAP estimates. Probabilistic model for linear regression and corresponding maximum likelihood solution. Link functyion. Connection between regularisation and Maximum A Posteriori estimate. Regularised linear models and corresponding priors to parameters.
Slides: PDF
Video: UTTV(2016) UTTV(2015)
Literature:
- Duda, Hart & Stork: Patter Classification pages 84-107
- Bishop: Pattern Recognition and Machine Learning pages 137 - 161
- Bishop: Pattern Recognition and Machine Learning pages 204 - 220
Complementary exercises:
- Bishop: Pattern Recognition and Machine Learning pages 173 - 177
- Bishop: Pattern Recognition and Machine Learning pages 220 - 224
- Practical comparison of various linear regression methods on data with different error distributions.
Free implementations:
- Built-in stats package in R:
glmanova. - LARS package in R
- Quantreg package in R:
rq