Day 15

Bayesian beginnings

With Charisse doing a phd which involves modeling with Bayesian networks, I'm starting to read up on this sort of thing myself (so that we can still talk to each other). I've been skimming her books and also getting a few myself from the kindle store (I seem to be able to get around to reading things if they are on my iPad). The idea seems quite simple so that is good for someone like me - I don't really get statistical type modeling really. But it seems to be all about directed graphs where the connections represent the existence of some sort of probabilistic relationships between the nodes being connected.

For example, say I have A and B where A can take on values 'good upbringing' or 'bad upbringing' and B can have values 'future male stripper' and 'future chief justice' then we might have a graph of the form
A --> B
Then perhaps have probabilities like
P(B=fcj | A=gu)
P(B=fcj | A=bu)
P(B=fms | A=gu)
P(B=fms | A=bu)
And these could be calculated as an output (think bayes theorem) given certain input data like probability of good upbringing etc.

Well that's my very early limited understanding of the idea. I'm sure you folk reading this can clear me up.

Students

So this all got me thinking when someone publishes a paper in this type of stuff, what do they write about in terms of results? For example when I write a paper with say PDEs for some biological application I present graphical output for example of spatial or temporal or both, solutions. Or I present an analytical result that might uncover a key parameter relationship. So what is the equivalent for w Bayesian network paper?

Then i got to thinking about students and research training. Perhaps sometimes, because of our (researchers) familiarity with what we are doing, we forget that the actual idea behind writing a paper and presenting results is not so intuitive for everyone. I think that's something I will definitely be keeping in mind in the future with my student supervision.