Reading Seminar on Algebraic Geometry and Singular Learning Theory
Time: 
Friday, 10:00  12:00 
Room: FR 6046 

Organizers: 
Summary
Singular Learning Theory is the study of singular parametric estimation, where naive application of classical learning and model selection methods like MaxLikelihood, Bayes Learning, AIC or BIC fails. As virtually all meaningful and practically relevant learning machines like Neural Networks, Mixture Models, Hidden Markov Models or Boltzmann Machines are singular, the analysis of their singular properties is of high practical relevance. Sumio Watanabe has developed generalizations of Bayes Learning Theory and Bayes Model Selection for the singular case; the aim of this seminar is the study of his work and its ramifications.
Prerequisites
Knowledge in Algebraic Geometry, Singularity Theory, Parametric Statistics and Bayes Estimation Theory is useful, but not necessary; all relevant basics will be discussed in the course.
Schedule
The topic list refers to the book Algebraic Geometry and Statistical Learning Theory, by Sumio Watanabe.
Date 
Topic 
Discussion leader 
20 April 2012 
Organizatorial Meeting and Overview on Singular Learning 
Duncan Blythe and Paul Larsen 
11 May 2012 
Chapter 1: Learning Theory and Singular Parametric Models 
Franz Király 
18 May 2012 
No Seminar 

25 May 2012 
Chapter 1: Singular Examples, Learning and Generalization Error 
Franz Király 
1 June 2012 
Chapter 1: Evidence and the Bayes quartet 
Paul Larsen 
8 June 2012 
No Seminar 

15 June 2012 
Chapter 1: The Zeta Function 
Duncan Blythe 
22 June 2012 
Algebraic Geometry Basics 
