Can ABC be Used for Model Selection? | Zendy

Xavier Didelot | Zendy; Rich Everitt | Zendy; Adam M. Johansen | Zendy; Daniel J. Lawson | Zendy

AI Assistant Blog Pricing

Home ZAIA Blog

Open Access

Can ABC be Used for Model Selection?

Author(s) -

Xavier Didelot,

Rich Everitt,

Adam M. Johansen,

Daniel J. Lawson

Publication year - 2011

Publication title -

nature precedings

Language(s) - English

Resource type - Journals

ISSN - 1756-0357

DOI - 10.1038/npre.2011.5955.1

Subject(s) - approximate bayesian computation , exponential family , computer science , model selection , statistic , selection (genetic algorithm) , curse of dimensionality , likelihood function , bayesian information criterion , computation , bayesian probability , artificial intelligence , statistics , mathematics , machine learning , algorithm , estimation theory , inference

Over the past ten years, Approximate Bayesian Computation (ABC) has become hugely popular to estimate the parameters of a model when the likelihood function cannot be computed in a reasonable amount of time. ABC can in principle be used also to perform Bayesian model comparison, but this raises the question of which summary statistic should be used for such applications. Here we present a general method for constructing a summary statistic that is sufficient for the model choice problem. We apply this construction to models from the exponential family. Unfortunately, in more complex models, our construct often results in statistics with too high dimensionality to use in ABC. We therefore discuss the possibility of applying ABC with non-sufficient statistics

The content you want is available to Zendy users.

Already have an account? Click here to sign in.

Having issues? You can contact us here

Accelerating Research