How would you design and handle training NN where the label is 1000++ bit binary (50% ones, 50% zeros). The number of labels can be small OR big in different situations.
My question: is such scenario well suited for NN ? If that seem like OK how will you design it.
For the sake of example let say I'm training the NN with MINST db. I.e. 10 labels.
Again the important thing is that the LABEL is with the specification I described. I can drop 50% ones, 50% zeros requirement if absolutely necessary,but the size of the binary have to stay.