Hello, I am looking for a good introductory book on Dynamic Bayesian Networks. I have experience with genetic algorithms but I want to branch out a little bit. I read the excellent "AI Techniques for Game Programming" and it was perfect because it had lots of examples and hand-holding along the way. Whenever I try to read tutorials online about Dynamic Bayes Nets it immediately turns into a bunch of math and I get completely lost. Are there any good books or resources for the "average" person? Thanks! :)

Unfortunately, Bayesian networks is a lot of math. Don't know if you will be able to find a book that doesn't treat it as such.

Well, it's math, but as long as you understand the underlying concepts in probability, it isn't especially difficult math. If you want a good book on Bayesian networks, you want this book. It is, to put it simply, The Book on bayesian networks. A bit expensive, so see if a nearby library has it.

While Judea Pearl's book IS considered a foundation tome in the area of graphical models for probabilistic reasoning, it isn't the most accessible work on the subject. For those not familiar with formal methods in mathematics you might get a bit lost in the detail. There are far better introductory and intermediate level books on the subject of Bayesian Networks and Dynamic Bayesian Networks (the latter being far more intricate and difficult to deal with than they're often made out to be). Here is a short list of excellent books...

Finn Jensen: Introduction to Bayesian Networks
(Very easy to read and accessible with little prior experience in probabilities or Bayesian methods)

Rich Neopolitan: Learning Bayesian Networks
(Rick actually did a lot of the early work on BNs and wrote a foundation book before Judea did... but his publishers didn't get it out in time and Judea's went down as the first definitive book in the field)

Kevin Korb & Ann Nicholson: Bayesian Artificial Intelligence
(This is a comprehensive book covering Bayesian techniques in AI and while it does provide a lot of mathematics, this is always accompanied by good explanations of the concepts and many clear examples.)


In the area of Dynamic Bayesian Networks you should try and get through Kevin Murphy's PhD thesis. He provides an excellent coverage of the background research, methods and problems associated with this approach to modelling stochastic processes using graphical methods.

If you want more information on DBNs, graphical models or probabilistic inference I'd be happy to help. It's a central theme in my research and I always love chatting about them! ;)

Cheers,

Timkin

Sorry to hijack this thread but it grabbed my attention instantly :)

@Timkin:

Where and what kind of research are you doing on Dynamic BN's? My colleague at work is doing his PhD thesis on applying DBN's to massive (huuuuge) datasets. Also our company currently uses static and dynamic BN's in the software we write (http://www.complex-adaptive-systems.com). In fact, our boss, Dr Anet Potgieter has several patents on agent-based BN's...

Look forward to hearing from you :)
Kurt

Quote: Original post by kurtkz
@Timkin:

Where and what kind of research are you doing on Dynamic BN's?

My earliest work was on efficient inference in DBNs and DDNs for continuous arbitrary (typically nonlinear, highly coupled) processes and drawing together the methods from nonlinear filtering with DBNs. Mostly I deal with DBNs for modelling physical processes and so I focus on approximate inference algorithms and issues caused by discretisation of time.

I have some former colleagues who did a LOT in the area of Bayesian techniques for data mining (basically the group at Monash I was a part of), so I've had a little exposure to the issues in that area... but didn't take a huge interest (I probably should have... apparently it pays pretty well in industry!) ;)

Cheers,

Timkin