Five years ago, I trained a GAN deep model to synthesize data based a tabular training data. Colleagues extended thus to also handle categorical and string data

Five years ago, I trained a GAN deep model to synthesize data based a tabular training data. Colleagues extended thus to also handle categorical and string data. Models trained on synthetic data performed well in withheld real data. I mention this as some evidence that deep models subsume conventional ML.