New Techniques for the Control of Multi-Agent Systems in Uncertain Environments
Venkat Anantharam, Amin Aminzadeh Gohari and Charles Bordenave
The main ideas in the project are (1) to tame the complexity of controlling multi-agent systems by a focus on aggregate behavior, i.e., not so much who carries out a given action as how many carry out a given action, and (2) to investigate the use of common randomness in the design of distributed systems.
The principal technique we propose for taming complexity by aggregation is the use of mean-field methods. These methods have proven well-suited to exposing the key qualitative features of statistical mechanical systems in physics but have so far not been much used for control. In our work we are developing a mean-field control theory to fill this gap. This is likely to be of particular interest in novel application areas of control such as control of chemical reactions, as well of use in the understanding of control of systems with a very large number of agents (e.g., sensor networks).
The investigation of the use of common randomness is currently focused on two areas: distributed game theory and information theoretic security.