Reasoning patterns

• Steps towards a model :

  1. Define the concepts/variables of interest
  2. Model the (causal) dependency structure between these variables
• (CPDs)

  Unconditional probabilities are associated with variables without parents.

  Conditional probabilities for variables that depend on parents variables.
• Each row in a table (of CPDs) is a conditoinal probability distribution
• Reading CPDs

  A) Unconditional probabilities

  B) Unconditional probabilities

  C) Conditional probabilities

  D) Conditional probabilities

  E) Conditional probabilities

  
• A probalistic query involes computing the Conditional Probability Distribution (CPD) $P(X | E = e ) =$$P(X | e)$

  "Given that we observed the specific values e of the variable E, what are the probailities for the different value combinations of the X"
• Causal Reasoning or Prediction 


  reasoning from causes to effects ("downwards" in the network)
• Evidential Reasoning or Explanation


  Reasoning from effects/obsedrvations to possible causes ("upwards" in the network)
• Intercausal Reasoning 


  a.k.a "Explaining Away", one causal factor gives us infromatin about another causal factor
• Reasoning with Bayesian Networks

  • A reasoning algorithm can dervie every consequence that follow from any cmobination of given information
  • No notion of cause and effect needed : dependencies in a network can be modelled in anrbitrary directions