Privacy-Preserving Federated Learning for Smart Home

With the advent of technologies such as internet of things, big data, and next generation communication systems, increasing amount of personal information is collected from devices that are typically used in home, such as thermostat, entertainment systems, and appliances.

The data generated from the smart homes can be fed to (distributed) AI/ML algorithms for smart decision making and more personalized recommender systems. For example, these  personalized recommender systems can help the home owners to reduce the electric bill, recommend entertainment content and healthcare related tips. 

While the concept of AI/ML is beneficial, availability and processing of this data also results in data privacy/security and concerns. These concerns will inhibit the home owners supporting the distributed learning or privacy-preserving AI/ML for personalized recommendation systems.

To address the requirements for state of art privacy preserving concerns, there is an immediate need for tools to (i) address such privacy/security and concerns and (ii) inform data owners about their data sharing decisions. As such, we propose to develop differential privacy-based utility- and privacy-preserving techniques to collect data, and build models using the collected data that results in unique implementation of privacy-preserving AI/ML model development in a smart home context.