The Data Minimization Principle in Machine Learning

Explore the concept of data minimization in machine learning through this 59-minute Google TechTalk presented by Ferdinando Fioretto. Delve into the challenges of implementing this principle, endorsed by global data protection regulations, in practical machine learning applications. Examine an optimization-based formalization attempt and its empirical analysis, revealing potential gaps between privacy expectations and actual benefits. Investigate the application of data minimization in high-stakes inference tasks, focusing on a sequential algorithm that identifies minimal attribute sets for accurate predictions. Gain insights into the importance of formalizing privacy legal principles for actionable deployment in the field of machine learning.