lecture 6: hidden variables and expectation-maximization
DESCRIPTION
Maximum Likelihood Estimation, Hidden and Latent Variables, Expectation-Maximization, EM for Naive BayesTRANSCRIPT
Machine Learning for Language Technology Lecture 6: Hidden Variables and Expecta6on Maximiza6on (EM)
Marina San6ni Department of Linguis6cs and Philology Uppsala University, Uppsala, Sweden
Autumn 2014
Acknowledgement: Thanks to Prof. Joakim Nivre for course design and materials
Repe66on: Baysian Approach to Classifica6on
MLE
Hidden and Latent Variables
Expecta6on-‐Maximaza6on (1)
Expecta6on-‐Maximaza6on (2)
Expecta6on-‐Maximiza6on (3)
Expecta6on-‐Maximiza6on (4)
Naive Bayes Revised
• Equa6ons are ok
EM for Naive Bayes
A Simple Example
Supervised Learning
Unsupervised Learning
First Guess
First E-‐Step
First M-‐Step
Second E-‐Step
Second M-‐Step
Third E-‐Step
Third M-‐Step
Fourth E-‐Step
Fourth M-‐Step -‐ Convergence
Uses of EM
The end