BNet.News
In This Issue: July/August 2005 
•   BNet.EngineKit 1.0 Released
•   BNet.Builder User Tip: Mutual Information
•   Charles River Analytics Presents BNet at UAI Conference
•   BNet.EngineKit User Tip: Mutual Information
•   Charles River Analytics Incorporates BNet.Builder into Their Behavior Modeling Tool
•   Publicize Your Belief Network Success Story
BNet.EngineKit 1.0 Released
BNet.EngineKit 1.0 was released in mid-July.

BNet.EngineKit is a developer toolkit for researchers and engineers to use to embed belief networks in software applications. Unique in its focus on clear APIs with the right functionality, BNet.EngineKit offers software developers who aren’t inference algorithm specialists a chance to use Bayesian networks without spending years learning about them.

EngineKit 1.0 includes learning functionality that allows you to learn your CPTs from data. Two easy-to-use algorithms learn CPTs when you have data for all nodes in your network (fully observed) and when you have missing data or hidden nodes (partially observed). The learning
algorithms use a standard interface for obtaining data, so you can also create your own implementation and provide data from any custom source.

You can find out more about Engine.Kit and belief networks on our Web site. You can also download a trial version of Engine.Kit and purchase the full version.

See BNet.EngineKit in Action

Charles River Analytics Presents BNet at UAI Conference
EDINBURGH, SCOTLAND – Charles River Analytics sponsored the 2005 Uncertainty in Artificial Intelligence (UAI) Conference, which took place in Edinburgh, Scotland from July 26-29, 2005.

Sean L. Guarino and Dan Grecu exhibited Charles River Analytics’s Bayesian Belief Network tool, BNet.Builder, at the conference. BNet.Builder is a desktop application for rapidly creating Belief Networks, entering information, and getting results. BNet.Builder is part of Charles River Analytics’s family of Belief Networking tools, including the recently released BNet.EngineKit. BNet.EngineKit is a developer toolkit for researchers and engineers to embed belief networks in software applications.

To read more about the BNet family of products, click here.

Charles River Analytics Incorporates BNet.Builder into Their Behavior Modeling Tool
Charles River Analytics incorporated BNet.Builder into their Modeling Individual Differences and Stressors (MINDS) project. MINDS employs a cognitive modeling approach and associated modeling tools that incorporate the effects of behavior moderators on decision-making behavior in skilled operational environments. MINDS is intended to enhance the performance of individual soldiers and small units by supporting the investigation of the effects of a range of moderating factors on cognitive performance, including personality and skill, as well as operational factors, such as fatigue and heat stress. The expected result is a suite of modeling tools that will allow US Army researchers to generate and evaluate specific hypotheses regarding operational performance in the small unit combat environment.

Charles River Analytics presented a paper on the MINDS project at the 14th Conference on Behavior Representations in Modeling and Simulation, Universal City, CA this past May.

To read more about the MINDS project, click here.

To read the abstract and/or request a copy of the paper, click here.

Moderated belief network
BNet.Builder User Tip: Mutual Information
BNet.Builder provides easy access to a form of sensitivity analysis called mutual information. The mutual information for a node (called the query node) in a Bayesian network is a value for every other node in the Bayesian network that represents how new evidence for the node would increase certainty in the beliefs of the query node. Using mutual
information, you can determine the node you should obtain evidence for to increase certainty the most in the beliefs of the query node. The higher the mutual information value, the more that the evidence for that node will increase certainty in the beliefs of the query node.

To display mutual information, first open the Mutual Information window using one of the following procedures:

- Right-click on a node and select Mutual Information
- Use the Ctrl+2 shortcut
- Select Mutual Information from the Window menu

When the Mutual Information window is open, you can select any node in the network to view its mutual information values.

Mutual Information
BNet.EngineKit User Tip: Mutual Information
BNet.EngineKit also provides easy access to mutual information values. The com.cra.bnet.engine.DecisionAnalysisTools class provides two methods for obtaining mutual information:

- getMutualInformation(BayesianNetwork network, DiscreteNode queryNode)
- getMutualInformation(BayesianNetwork network, DiscreteNode queryNode, double threshold)

Both methods take a Bayesian network and a query node and return a java.util.Map object that maps node names to mutual information values. The second method also takes a threshold value that is used to filter values from the returned Map (only values above the threshold will be included).

The following code sample obtains mutual information values for a query node and prints them out:

BayesianNetwork network = ...;
DiscreteNode node = ...;
Map values = DecisionAnalysisTools.getMutualInformation(network, node);
System.out.println("Mutual Information values for " + node.getName());
System.out.println(values);

Publicize Your Belief Network Success Story
Want to show off your applications of belief networks to others using them in their research? Send us your success stories. We would like to include at least one in each edition of BNet.News.

Send us your success story

BNet News is published by Charles River Analytics, the company that brings you BNet.Builder and BNet.EngineKit.  View our Privacy Policy