Privacy-preserving inference and decision-making with IoT data

Principal Investigator: Osman Yagan, Associate Research Professor, Electrical and Computer Engineering, College of Engineering

Co PI: Gauri Joshi, Assistant Professor, Electrical and Computer Engineering, College of Engineering

We have received funding from the Carnegie Bosch Institute for Privacy-Preserving Inference and Decision-Making with IoT Data. In the age of Internet-of-things (IoT) and edge computing, the state of complex systems, e.g. manufacturing/industrial systems, connected mobility systems, smart homes or personal health networks, can be measured via various data-collection mechanisms such as distributed sensors, vision systems, LiDAR and crowd-sourced workers. These data are an essential component of intelligent systems that perform autonomous decision-making and inference e.g. smart industrial robots and digital assistants that perform user preference based cognitive tasks. However, the ubiquitous data-collection can reveal sensitive information about individuals and violate their privacy. It is envisaged that the commercial adoption of such systems will be constrained by the privacy issues, in both regulatory domain (e.g. General Data Protection Regulation) and from the users’ trust perspective.

Our research goal is to enable statistical inference, and learning systems without compromising individual privacy. To achieve this goal we will pursue: 1) the design of data-collection mechanisms that protect individual privacy, while still providing useful information about the system as a whole, and 2) provably optimal techniques to combine information collected from heterogeneous sources, and 3) algorithms to sequentially obtain measurements in order to minimize the cost of data-collection. We plan to evaluate and deploy the proposed techniques on real-world datasets obtained through our collaborations with industrial partners and National laboratories. The research will not only enable privacy-preserving and efficient planning, learning and inference for IoT systems, but will also contribute to fundamental advances in the field of statistical estimation and multi-armed bandit algorithms.