Markov Networks
Graph structure encodes independence assumptions:
Chris
conditionally independent of
Eve
given
Alice & Dave
Chris
Dave
Eve
Alice
Betty
ABC
Compatibility
f
(A,B,C)
One key difference between bns and mns is that in bns we model dependence of each variable on others, while
in mns, we model interactions of sets of variables.
This buys us the freedom from acyclicity constraint at the
expense of difficulty of interpreting the parameters.
It handles symmetric interactions, where cause and effect cannot be identified.