Catching knowledge in neural nets