Thursday, March 5, 2009

Ancient Hinduism and Bayesian inference

What is common among the two?

1. Bayesian inference accommodates and absorbs other schools of inference, much like Hinduism did to invading tribes.

2. In both Bayesian thought and (some schools of orthodox) Hinduism, the underlying variable (fundamental truth) is non-deterministic.


Pradeep Mohan said...

Seems like you have become a bayesian by religion!

Pavan said...

Haha. Actually, I first heard the term 'Bayesian religion' used by an anit-Bayesian dude from San Diego, who claimed that 'the Bayesian religion' has corrupted/subsumed all statistical inference. Look here for more.

Pradeep Mohan said...

yeah yeah!! I actually met and took the class of a bayesian fanatic! worst of all this guy is going to be in my committee... because he is the only spatial statistician.... in my univ..

Pavan said...

I've taken two courses dealing with the books by Andrew Gelman and Chris Bishop respectively. I think together they cover most of the current thought on Bayesian inference.

Don't worry about the committee and shit! It's really simple. The core ideology is this: More freedom and a convenient framework to selectively weight observed data and prior knowledge.

The rest is in the engineering. More specifically, 3 'how' questions.

How to use the existing machinery of linear algebra/sampling methods/classical statistics etc. to do Bayesian inference?

How to situate Bayesian inference in the larger context of statistical testing? [confidence intervals, error bounds].

And perhaps the most important: how to select a useful observation model and useful priors based on the phenomenon at hand and the tools available.