Source: Joana Tirana et al., "The Role of Compute Nodes in Privacy-Aware Decentralized AI" in 6th Int'l Workshop on Embedded and Mobile Deep Learning (July 2022)
The Main Idea.
- Modern mobile and IoT devices store tremendous amounts of data in a nearly continuous fashion given our devices tend to be "always on" and even, always on us (e.g., Apple watch tracking sleep behaviors). Machine learning models are being applied to this vast data repository, but in "many cases, these ML models are produced (trained) using personal, potentially privacy-sensitive data" (Id. at 19).
- Working on decentralizing compute can be a way to preserve data-level privacy while increasing scalability vs. centralized AI/ML approaches.