Fun coincidence (or maybe not?), I just stumbled upon the following paper on arvix titled "Differentially Private Generative Adversarial Network" https://arxiv.

Fun coincidence (or maybe not?), I just stumbled upon the following paper on arvix titled "Differentially Private Generative Adversarial Network" https://arxiv.org/pdf/1802.06739v1.pdf https://arxiv.org/pdf/1802.06739v1.pdf Its Related Work section also references some works on "differentially private learning in neural network". While the paper does not discuss directly OP remark about how differential privacy could help generality in machine learning, it shows that at least some connections exist between the domains.