Bayesian Nets

• Statistics may be the best form of machine learning. It not only gives you answers, but it also gives you the likelihood that the answer is correct (power).
• We don't use it for all learning, because we often don't understand the statistics of the problem well enough.
• P(L|W) = .75
  P(L|M) = .15
  P(W) = P(M) = .5
  What's P(W|L)?
• P(W|L) = P(L|W)P(W)/P(L)
  P(W|L) = P(L|W)P(W)/P(L|W)P(W)+P(L|M)P(M)
  P(W|L) = .75*.5/(.75*.5)+(.15*.5) = 5/6 ~= .83
• That's an exanmple of Bayes rule
• You can chain Bayesian equations together to get a Bayesian Net.
• You can induce the values from training data.