354 14 Bayesian Networks
Figure 14.8 Modifying a BN by reversing the direction of a dependency when two
Boolean nodes are related by subsumption.- uncertainty analysis,
- consistency checking.
- A BN that fails a test must be modified.
- BN modification can be used as a normal part of BN development.
- BN modification operations have been identified and classified, and guide-
lines for when to apply them have been developed.
14.4 Exercises
- In the diagnostic BN in figure 14.1, one can use either a temperature mea-
surement or a patient’s perception of a fever to diagnose influenza. Al-
though these two measurements are a priori independent, they become
dependent when one observes that the patient has the flu or a cold. In
statistics this is known as Berkson’s paradox, or “selection bias.” It has
the effect that a high temperature can reduce the likelihood that a patient
reports being feverish and vice versa. Compute the JPD of the PF and T
nodes in this BN given the observation that the patient has influenza. - Compute the probability that a patient has influenza using temperature
measurements. For example, try 37◦,38◦,39◦,and40◦C. These are all
(in theory) exact measurements. In fact, a thermometer, like all sensors,
can only give a measurement that is itself a random variable. Compute
the probability of influenza given a temperature of 38.40◦C, normally
distributed with standard deviation 0.20◦C.