I am analyzing whether Neural Networks will be able to handle our
current classification problems, with varying input and output data-
types (not necessarily all types being known at time of NN creation).
Do any existing NN implementations allow this, wherein we can add
input neurons to an existing NN. Let's say I have trained a NN to
learn about 2 input data types (d1 and d2) with outputs o1 and o2
Is it possible to further train this NN to now handle additional input
data type (d3) and additional output o3 nodes. Is it theoretically
possible and whether any implementations handle it (for e.g. FANN
etc)?
Thanks
Samir
