Answered by:
Confused: Bayes Point Machine vs Bayesian Network vs Naive Bayesian (Migrated from community.research.microsoft.com)
Question

shankarooni posted on 01192010 4:07 PM
I get confused about these terms, so please help me check my understanding:
Bayes Point Machine (BPM) is an algorithm to inference a Bayesian Network (both) naive or believe (T/F)?
EP & VMP are algorithm which work/ help BPM...(T/F)
We do not have a Bayesian network in case of naive as all variables are independent from each other (T/F)?
and lastly
An example of naive Bayesian is the BayesPointMachineExample (T/F)
Thank you...
Friday, June 3, 2011 5:32 PM
Answers

shankarooni replied on 01242010 3:46 AM
Dear John, David
Thank you very much! That helped a lot ,
Got a lot of reading to do...
Thanks again.
Last question, What to use when my feature vectors contain discrete values (like true/false) and not continuous values.
 Marked as answer by Microsoft Research Friday, June 3, 2011 5:32 PM
Friday, June 3, 2011 5:32 PM
All replies

DavidKnowles replied on 01202010 7:13 AM
Hi
EP and VMP are algorithms to perform inference on general Bayesian networks (which I would rather call models). Infer.NET lets you specify the particular Bayesian model that you want, and automatically generates and runs code to perform either EP or VMP on your model. The NB and BPM models are both examples of "Bayesian networks", and are both used for classification tasks. Naive Bayes is a very simple model where each feature has a distribution which depends on the class the sample belongs to, but is independent of all other features. The BPM is closely related to a Support Vector Machines.
Good background reading on these topics would be Chris Bishop's book: http://research.microsoft.com/enus/um/people/cmbishop/prml/
Or my supervisor's course (particularly Chapters 4, 5, 6, 10 and 11): http://mlg.eng.cam.ac.uk/teaching/4f13/0809/
I hope that helps!
David.
Friday, June 3, 2011 5:32 PM 
John Guiver replied on 01222010 5:50 AM
Bayesian belief networks are a family of graphical models which have discrete variables.
A Bayes Point Machine is a specific type of graphical model which maps observed (typically nondiscrete) feature vectors to discrete class variables. A description is given in the Multiclass classification with BPMs section of the user guide. .
Naive Bayes is a generative type of model in which features are considered independent conditioned on a class variable. This is a very simple form of classification based on Bayes theorem  a good description is given in the wikipedia article.
EP and VMP are general purpose inference algorithms that can be applied to any graphical model including Bayesian networks, BPMs, and naive Bayes.
John
Friday, June 3, 2011 5:32 PM 
shankarooni replied on 01242010 3:46 AM
Dear John, David
Thank you very much! That helped a lot ,
Got a lot of reading to do...
Thanks again.
Last question, What to use when my feature vectors contain discrete values (like true/false) and not continuous values.
 Marked as answer by Microsoft Research Friday, June 3, 2011 5:32 PM
Friday, June 3, 2011 5:32 PM