Privacy-Preserving Distributed AI for Smart Homes

Privacy-Preserving Distributed AI for Smart Homes

Distributed AI (D-AI) is enabling rapid progress on smart IoT systems, homes, and cities. D-AI refers to any AI system with discrete AI subsystems that can be combined to create ensemble-intelligences. We introduce a Privacy-Preserving IoT Cloud (PPIC) optimized to host D-AI applications on the edge.

Today’s smart homes, cities, and infrastructures are often shoe-horned into cloud-computing architectures created for the Web. This is risky: any system that uploads sensitive data into a data warehouse concentrates secrets. We favor a Privacy-Preserving IoT cloud (PPIC) that decentralizes intelligence, running distributed AI (D-AI) logic at the edge while preventing sensitive data from leaking.

The PPIC is an execution framework that marries edge-computing (computation near or on IoT devices) with a scalable distributed-AI computing model. Our PPIC framework can host third-party apps, supports multitenancy, and offers scalable services for tasks like object storage. We depart from the cloud model by replacing the centralized datacenter with a decentralized graph, so that sensitive data is never uploaded to the cloud. Smart applications run at the edge, enabling better performance, while policy-driven object filtering blocks unauthorized sharing of sensitive information.

The PPIC is a natural fit to D-AI in smart homes, multi-home developments and smart cities. A PPIC can safeguard privacy while providing scalability and cost efficiency. Here, we illustrate this with a range of tasks: operating a light switch, contextualized voice recognition, and fall detection for the elderly. PPIC can also be used at larger scale, for example to support city-wide coordination of emergency responders during a weather or wildfire emergency, but this paper focuses on functionality already in active use.

Visit Caspar Technology for more details.