next up previous
Next: Designing Bayesian Networks 2 Up: MLA_Exercises_160106 Previous: Improving Performance by Feature

Designing Bayesian Networks 1 [3 P]

In a nuclear power plant you have an alarm that senses when a gauge, which measures the core temperature, exceeds a given threshold. Consider the Boolean variables $ A$ (alarm sounds), $ F_A$ (alarm is faulty), and $ F_G$ (gauge is faulty). Assume also that the core temperature $ T$ and the measurement of the gauge $ G$ can take on only two possible values normal and high.
a)
Draw a Bayesian network for this domain, assuming that failure of the gauge is independent of the core temperature. Explain every step of your construction.
b)
Construct the conditional probability tables for the Bayesian network, using the following information: The gauge shows the correct temperature 99% of the time, when it is working, but only 70% of the time when it is faulty. The alarm always works, unless it is faulty. If it is faulty, it goes off 5% of the time, no matter what the gauge shows. The gauge is faulty 2% of the time, the alarm is faulty 0.5% of the time.
c)
Suppose we know that the gauge is working and the alarm sounds. Calculate the probability that the core temperature is too high as a function of the prior probability of the core temperature being too high $ P(T=high) = t$ .


next up previous
Next: Designing Bayesian Networks 2 Up: MLA_Exercises_160106 Previous: Improving Performance by Feature
Pfeiffer Michael 2006-01-18