The objective of this project is to develop a unified computation framework for distributed adaptive sensing and control in stochastic multi-agent networks with either static or dynamic states, coupled with model uncertainties. In such a fully distributed information processing and control paradigm, collaborative communication and participation at the agent level is explored to build efficient scalable architectures for real-time pervasive decision making with provable performance guarantees. The distributed methodologies are amenable to real-time implementation and adaptation over parametric switching and uncertainties, through proper exploitation of locally sensed data. The resulting mixed-time-scale information and cyber/physical dynamics are analyzed with new novel techniques in stochastic analysis, with special considerations over the associated non-Markovian properties.
Intellectual merit: The intellectual merit is lying in the fact the proposed distributed adaptive framework involves a complex and subtle closed-loop interplay among stochastic learning, estimation, and communication, where the proposed approach leads to the novel mixed-time-scale sequential stochastic schemes of the consensus + innovation type. The dynamics of these procedures are non-Markovian, for which a completely new and transformative systematic approach is proposed to conduct the system performance analysis.
Broader impacts: The broader impacts are multi-fold. The results will enable efficient cyber-physical system operations, and will provide new methodologies for advanced dynamic system analysis. The proposed concurrent education program addresses the undergraduate and graduate education issues in multiple aspects: teaching, training, and mentoring, with outreach to the community and under-represented minority groups.