etc.
The learning rule for a logistic DAG is:
With replicated weights this becomes:
h2
v2
h1
v1
h0
The derivatives
for the recognition
weights are zero.
v0