GeNIe Tutorials: Tutorial 10 - Using control values
From DSL
Objective of the tutorial: To learn how to control values in networks.
Estimated Time: 10 Minutes
At the end of the tutorial you will be able to:
- Use control values.
In causal probabilistic models, there is an additional class of inference problems: 'Predicting effects of external intervention'. In the context of Bayesian networks, computing the effect of observation amounts to belief updating after setting evidence for the observed network variables. The effect of intervention, on the other hand, is a change in the network structure, related to external manipulation of the system modeled by the network, followed by setting the values of the manipulated nodes and updating beliefs.
We will explain control values with the help of the following example:
Consider a model for the operational status of a "Command Center." The Command Center depends on the status of "Communications" and Radar. Radar depends on the "Antenna Structure" and the power supplied by the "Generator." Communications rely on the power supplied by Generator. The Generator relies on "Fuel Supply" to generate power.
We have created a Bayesian network that models these causal relations.
This network is saved as "ControlValue.xdsl" in the Example Networks folder.
Load the Tutorial10.xdsl network into GeNIe.
If you need help in loading a network file into GeNIe, see Tutorial 5 : Loading and Saving Models.
Once the network loads, you will see the following diagram in the Graph View.
To learn how to build Bayesian Networks like the one above, See Tutorial : Building a Bayesian Network.
Each node in this network has two states: Damaged and Operational.
Now suppose the model is an enemy's command center and the decision maker's objective is to disrupt the enemy's communications. We can act on the communications by, for example, jamming its outgoing signal with noise. This can be viewed as an external intervention that results in Communications not working, i.e. essentially setting the value of variable Communications to state Damaged.
Communications will be Damaged regardless of the values of other variables in the network (and, in particular, its parent variable, Generator).
Control Values are used to model such type of situations.
Let us control the value of Communications node to Damaged.
- Right click on Communications node and select Control Value from the Node Popup menu that appears.
- Select Damaged from the Control Value submenu.
This action has the following effect:
- 1. Since a controlled node does not depend on the value of its parents, there is a temporary change in the network structure. GeNIe shows this by dimming the arc connecting the parent nodes to the controlled node. In our case, the arc connecting Generator and Communications is dimmed as shown below:
- 2. Now that we have controlled the value of the Communications node, GeNIe sets the value of the controlled node to the value imposed by the manipulation. So the value of Communications is set to "Damaged." To indicate that the node is controlled GeNIe displays the
status icon on the node. Notice that the Communications node has the status symbol.
To inspect the effects of intervention, we will need to update the model. After updating the model you can view the values of each node.
Notice that the intervention only changes the posterior probabilities of the descendants of the controlled node. The control value operation is not available for those nodes that have observed or manipulated descendants.
Next Tutorial:
