Hello!
How did you solve the problem of disbalance between positive and negative samples for binary CNN's training?
In other words, if there are only 100 samples of 16 y.o. and 50k+ samples -- other ages, how did you choose samples for training first binary CNN (which returns an answer for "Is this person older than 16?").
Thank you.