Learning Bayesian Networks
- Suppose structure known, variables partially
observable. For example we observe (ForestFire, Storm,
BusTourGroup, Thunder), but not (Lightning, Campfire)
- We can use a method similar to training neural network
with hidden units.
- Specifically, we will use gradient ascent.
- We want to converge to network $h$ that (locally) maximizes $P(D\,|\,h)$
José M. Vidal
.
30 of 39