Abstract | ||
---|---|---|
This paper has been accepted for presentation at the twentie th NIPS (Neural Information Process- ing) Conference, December 2007, Vancouver, Canada. The fina l version of the paper, to appear in the conference proceedings in 2008, will be slightly differ ent from the present one. Abstract Bayesian model averaging, model selection and their approximations such as BIC are generally statistically consistent, but sometimes ach ieve slower rates of con- vergence than other methods such as AIC and leave-one-out cross-validation. On the other hand, these other methods can be inconsistent. We identify the catch-up phenomenon as a novel explanation for the slow convergence of Bayesian meth- ods. Based on this analysis we define the switch-distributio n, a modification of the Bayesian marginal distribution. We prove that in many situations model selection and prediction based on the switch-distribution is both con sistent and achieves op- timal convergence rates, thereby resolving the AIC/BIC dilemma. The method is practical; we give an efficient implementation. |
Year | Venue | Keywords |
---|---|---|
2007 | Reviews of Modern Physics | leave one out cross validation,convergence rate,information processing,model selection |
Field | DocType | Citations |
Convergence (routing),Mathematical optimization,Bayesian inference,Computer science,Bayesian average,Model selection,Bayesian statistics,Bayesian probability | Conference | 11 |
PageRank | References | Authors |
0.83 | 8 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Tim van Erven | 1 | 218 | 17.39 |
Peter Grunwald | 2 | 113 | 11.40 |
Steven de Rooij | 3 | 73 | 10.16 |