[screye]

Michael Jordan in a recent seminar talked about how they were developing a trade-off parameter that modelled privacy. By treating the trade-off parameter as a knob, you can control the trade-off between data quality and privacy compromise.

Similarly Google has started using methods like federated learning to learn from ML models without ever sending your data to the cloud.

The idea is certainly gathering traction, and it will be very interesting where this is going to from here.