Arc reversal is the process of changing the direction of a link between two nodes in a Bayesian network.
When a link is reversed, the overall distribution for the network is maintained (in other words, queries will return exactly the same values). In order to do this, some node distributions will be updated, and new links may need to be added.
Node absorption is the process of removing a node from a Bayesian network, while keeping the joint distribution of the remaining nodes the same.
To perform Node absorption to a node A, follow the following steps:
- For each child C of A, apply Arc reversal to the link A -> C.
- Remove A from the network (or remove all incoming links to A).
Note that, as with Arc reversal, node absorption may induce new links, in order to maintain the joint distribution of the un-absorbed nodes.