Fair Acquisition of Private Data in Multi-agent Systems

The vast amount of data generated by individuals in various aspects of life, from media to

healthcare and transportation, has been instrumental in solving complex problems across different domains. For example, many machine learning algorithms rely heavily on data that is provided directly by users. Moreover, in many societal scale engineered systems, collection of user data can enable a more safe and efficient system operation.

For instance, fine-grained power consumption data communicated through smart meters has enabled many technological advances in smart power systems. This NSF project explores how platforms can fairly acquire users’ data given their privacy requirements and their contributions to the performance of the distributed system, and how to optimally utilize such private data markets for safe and efficient operation of distributed learning and control systems.

This project’s intellectual merit lies in its innovative contributions, including the development of a novel framework for privacy-aware and fair data acquisition that integrates concepts from machine learning, optimization, and game theory. The broader impact of the project includes enhancing societal trust in AI-based technology through enabling fair and private data acquisition.

Further, the project engages undergraduate students in research and has outreach activities involving pre-college students.

The project’s goal is to design fair data acquisition mechanisms alongside control and learning algorithms that incentivize strategic agents to contribute the appropriate share of private data, ensuring efficient and safe operation of distributed systems with critical constraints, such as those found in autonomous driving and power systems. We propose a mathematical setting where users exchange their data with a platform for payment or services, considering varying privacy requirements, and focus on creating fair incentives for distributed learning and inference applications within this framework.

Building on this static analysis, we focus on dynamical systems which involve continuous data collection and feedback loops, introducing temporal correlations that complicate information leakage control and incentive design for data acquisition. We formalize these challenges as research questions, focusing on designing adaptive incentive methods for such dynamic systems.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.