Perceptual learning viewed as a statistical modeling process - Is it all overfitting?