back to Ockham’s razor

“All in all, the Bayesian argument for selecting the MAP model as the single ‘best’ model is suggestive but not compelling.”

Last month, Jonty Rougier and Carey Priebe arXived a paper on Ockham’s factor, with a generalisation of a prior distribution acting as a regulariser, R(θ). Calling on the late David MacKay to argue that the evidence involves the correct penalising factor although they acknowledge that his central argument is not absolutely convincing, being based on a first-order Laplace approximation to the posterior distribution and hence “dubious”. The current approach stems from the candidate’s formula that is already at the core of Sid Chib’s method. The log evidence then decomposes as the sum of the maximum log-likelihood minus the log of the posterior-to-prior ratio at the MAP estimator. Called the flexibility.

“Defining model complexity as flexibility unifies the Bayesian and Frequentist justifications for selecting a single model by maximizing the evidence.”

While they bring forward rational arguments to consider this as a measure model complexity, it remains at an informal level in that other functions of this ratio could be used as well. This is especially hard to accept by non-Bayesians in that it (seriously) depends on the choice of the prior distribution, as all transforms of the evidence would. I am thus skeptical about the reception of the argument by frequentists…

Laplace great⁶-grand child!

Looking at the Family Tree application (I discovered via Peter Coles’ blog), I just found out that I was Laplace’s [academic] great-great-great-great-great-great-great-grand-child! Through Poisson and Chasles. Going even further, as Simeon Poisson was also advised by Lagrange, my academic lineage reaches Euler and the Bernoullis. Pushing always further, I even found William of Ockham along one of the “direct” branches! Amazing ancestry, to which my own deeds pay little homage if any… (However, I somewhat doubt the strength of the links for the older names, since pursuing them ends up at John the Baptist!)

I wonder how many other academic descendants of Laplace are alive today. Too bad Family Tree does not seem to offer this option! Given the longevity of both Laplace and Poisson, they presumably taught many students, which means a lot of my colleagues and even of my Bayesian colleagues should share the same illustrious ancestry. For instance, I share part of this ancestry with Gérard Letac. And both Jean-Michel Marin and Arnaud Guillin. Actually, checking with the Mathematics Genealogy Project, I see that Laplace had… one student!, but still a grand total of [at least] 85,738 descendants… Incidentally, looking at the direct line, most of those had very few [recorded] descendants.