There's a lot of goodness in this one paper (14 page PDF). For one thing, it discusses 'federated learning' (FL), which is "a decentralized machine learning (ML) paradigm in which a global model is trained collaboratively across multiple participants (e.g., smartphones, hospitals, or institu-tions), without exchanging raw data." So much to think about here. But the result of this study is a set of great insights into the concept, or should I say concepts, of fairness. There are many different types of fairness, they vary across contexts, and they aren't interchangeable. See the diagram for more detail. The result is a harm-based model which takes into account the various forms of fairness implicated in FL environments which "highlights a disconnect between how fairness is defined in research and how harms manifest in practice." More from the eighth AAAI/ACM Conference on AI, Ethics, and Society.
Today: Total: [Direct link] [Share]
