Machine learning approach that trains neural networks to approximate posterior distributions by learning from simulated data.