Iconic Training and Effective Information: Evaluating Meaning in Discrete Neural Networks