Hypothetical example, say I wanted:
P(gender,ethnicity|age,hair); so that the input would aligned to a trained dataset of:
(gender,ethnicity,age,hair) => hat bought.
What approach is 'best' for computing ~gender and ~ethnicity given age,hair; in order to predict the hat bought?
The processing of the
inputs => hat can be done/learned offline whereas infering the missing input values shall be done online. The results of the online pass shouldn't be stored in the network.
FYI: I am considering two Recurrent Neural Networks one for each problem.