Log in

Graphical Models of Probability
Bayesian Networks, Decision Networks, and Probabilistic Relational Models
Recent Entries 
Hi, I have what I assume is a pretty basic technical question y'all might be able to give me some pointers for:

Assume for simplicity all variables can have one of some finite number of discrete values,

Given the value/state of some set of N variables Bn (n = 1..N), and pair-wise relationships to some variable C (i.e., given P(C=x | Bn=y) for all x, y, and n), I want to estimate P(C=x | B0..BN) for all x -- i.e., the distribution over C's possible values given the values of all the B's.

Caveat is: no assumptions whatsoever about the relationship between the B's -- some might be entirely coincident and others completely independant.

First, I'm curious if there's any generally accepted best-approach to blending independant indicators in the absence of any further information (though it seems trivial to show that any one solution can be horribly wrong for certain examples).

More, I'm interested in the case where you have a little more information than the above: where you can train up your values by example, so you actually can witness the B's behaving together (i.e., where you are given a training set that has all the B and C values for each example), but where you are still limitted to keeping only one (or a few) values for each Bn -- i.e., only pairwise information between B and C, and no data specific to combinations of B's.

On the surface, this looks like a natural application for a simple perceptron (one-layer, trivial back-prop), but I am wondering if there is something qualitatively similar which is more statistically based rather than mean-squared-error based. (I.e., I would like to maximize P(C|B,M) where M is my model which picks a value for C at random based on the probability distribution inferred from the observed Bn's.)

Don't assume I already know the obvious -- if I'm just describing some algorithm everybody learned as a child, please provide a pointer to it. :)

12th-Dec-2004 06:07 pm - Books
Hello People

For the beginning I want to apologize for my English I know is poor but i hope you can understand me

I'am student of University Of Science And Technology (AGH) in Cracow (Poland). I'am working about thesis that is about "Bayesian Networks Used In Medicine (nosoconial infections)". We are tying build prototype expert system's model. I want to buy a book about topics like "Creating bayesian networks from data", "spanning tree and other algorithms - implementation and examples". I need "good" book about basics and many examples. Could somebody recommend well - written book. I found couple on amazon.com but i don't know which i should buy. There is no Polish literature about Bayesian Networks :(. Please Help

19th-Oct-2004 01:35 pm(no subject)
(Cross posted to java_dev and algorithms.)

I just wanted to let you all know that I have started a community called compscibooks that may be of interest to bayesnets people. This is a discussion community for computer science books: textbooks, monographs, even fiction featuring topics in computer science. Anyone who would like to read or participate in discussions on CS books is welcome.

I would love to see reviews in the community, and will cross-post my reviews of CS1, CS2, graphics, AI, and other textbooks there. Topics include, but are not limited to, the list of interests for compscibooks. One set of upcoming reviews features six current Java books for a first course in computer science, which I am looking at with regards to a hypothetical overhaul of our CS program.

BTW, any comments on the books I have slated for review, suggested additions to the list for graphical models of probability, or suggested criteria would be welcome. I'll take recommended criteria under consideration and any comments into account after I complete my reviews.

Thanks in advance for comments, and I hope to see you over in the community.

William Hsu
Assistant Professor of Computing and Information Sciences
Kansas State University
3rd-Oct-2004 05:56 pm - BNJ 3.1
has been released and is available on the Sourceforge project page, at Files on project bnj

From now until I finish it, I will be working on the user manual.
1.	Basic Probability Theory
	a) Introduction
	b) Axioms
	c) Notation and Definitions
	d) Theorems
	e) Bayes rule
2.	Basic Graph Theory
	a) Definitions
	b) Mathematical Models
	c) Implementation Notes for BNJ
3.	Introduction to Bayesian Networks
	a) Graph Theory + Probability Theory = Bayesian Network
	b) Formal definitions
4.	Modeling with Bayesian Networks
	a) simple probability queries
	b) making the simple, complex
5.	Influence Diagrams
	a) Decision Theory + Bayesian Network = Influence Diagram
6.	Modeling Influence Diagrams
7.	Using the BNJ 3.1 GUI
	a) add nodes
	b) add edges
	c) change domains
	d) fill in conditional probability tables
	e) use inference
8.	Using BNJ 3.1 from command line
	a) inference
	b) dynamic Bayesian network unrolling
9.	Implementation Notes
	a) CPF
	b) Interfaces
	c) The Core Specification
10.	References
23rd-Sep-2004 02:11 pm - Dynamic Bayesian Networks
see the sexiness of DYNAMIC BAYESIAN NETWORKSCollapse )

... they exist in the next release of BNJ 3.1
13th-Sep-2004 11:09 am(no subject)
Hi, I currently use fuzzy decision tree for a project but I am interested in converted it to Baysian net to see if Baysian net can come up with a better prediction than the fuzzy tree. I am looking for papers or resources on how this can be done, e.g converting for decision tree to Bayes net and vice versa, also how they are directly compared to each other. If any one has any suggestion , please let me know. Thank you.
28th-Jul-2004 02:00 am - BNJ v3.0 Release and Website
The BNJ Dev Team is pleased to announce a new website dedicated to screen shots and movies of the new BNJ v3.0 in action! Find our new website here:


An example of one screen shot (Better quality images available for download):

Thanks for the support all, and look for our new alpha release on August 1st!

The BNJ Dev Team is pleased to announce the upcoming first alpha release of BNJ v3.0a.

Here is a screen shot from the animation of the junction tree (aka Lauritzen-Spiegelhalter or Hugin) algorithm:


William Hsu
25th-Jul-2004 04:52 am - New home for BNJ
BNJ has moved from:




William Hsu
2nd-Jun-2004 02:48 pm - Welcome zengeneral
Please welcome zengeneral to the bayesnets development team. He joins silpan and Julie Thornton as part of a pretty vigorous development effort on BNJ v3!

Edit, Fri 02 Jul 2004: Please welcome Andrew King to the KSU KDD group and the dev team as well, and add masaga (a KDD group member since 2001) to the list of BNJ dev team members with LJs!

William Hsu
This page was loaded Feb 24th 2017, 10:40 am GMT.