Machine Learning with Differential Privacy
Anand D. Sarwate · 2024
In this chapter we take up the problem of machine learning for private or sensitive data. The phrase “privacy-preserving machine learning” can refer to myriad models for privacy and learning. Machine learning is a term that metastasized to encompass a large variety of approaches to the problem of inferring structure in data. While many “classical” methods in statistics have been rebranded as “machine learning”, a useful distinction is that the latter places slightly more emphasis on the computational or algorithmic aspects of the inference problem. Many machine learning methods also attempt to be “distribution-free” in the sense that they try to make very few assumptions on the model generating the data.