inference_in_bayes_nets.txt

Bayes networks define probability distributions over graphs of random variables.

To find how many probability values are required to specify a Bayes Network, identify the number of input variables K for each node(i) and sum 2^(K(i)).

Complex Bayes network representations offer considerable advantage over naive unstructured join representations. 

D- Separation: used for defining independence based on the graph structure and known variables.

Inference in Bayes Nets: Nodes can be Query (want to find), Evidence (know these) or Hidden (not Query or Evidence).

Can ask a direct question or find the most likely explanation (what combination have the highest probability).


Variable elimination: can be more efficient than having to perform the whole enumeration.

Likelihood Weighting can accommodate for sample rejection.

Gibbs Sampling

Monty Hall: Updating probabilities after new information (1/3 -> 2/3)