PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

On Recurrence Formulas for Computing the Stochastic Complexity
Tommi Mononen and Petri Myllymäki
In: 2008 International Symposium on Information Theory and its Applications, 7-10 Dec 2008, Auckland, New Zealand.

Abstract

Stochastic complexity is a criterion that can be used for model selection and other statistical inference tasks. Many model families, like Bayesian networks, use multinomial variables as their basic components. There now exists new efficient computation methods, based on generating functions, for computing the stochastic complexity in the multinomial case. However, the theoretical background behind these methods has not been been extensively formalized before. In this paper we define a bivariate generating function framework, which makes the problem setting more comprehensible. Utilizing this framework, we derive a new recurrence relation over the values of a multinomial variable, and show how to apply the recurrence for computing the stochastic complexity. Furthermore, we show that there cannot be a generic homogeneous linear recurrence over data size. We also suggest that the presented form of the marginal generating function, which is valid in the multinomial case, may also generalize to more complex cases.

PDF - PASCAL Members only - Requires Adobe Acrobat Reader or other PDF viewer.
EPrint Type:Conference or Workshop Item (Paper)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Computational, Information-Theoretic Learning with Statistics
ID Code:4188
Deposited By:Tommi Mononen
Deposited On:19 January 2009