A new approach to PCGML uses discriminative models to improve design control.
problem Fundamental tension between detailed training examples and return on investment.
method Discriminative models trained on positive and negative examples.
result Demonstrates new mode of control for learning-based generators.