Everhart Lecture

The smart grid, the "new" Internet, automated highways and the human sensorimotor control system: all of these massively complex systems can be described in terms of a large number of physically distributed subsystems interacting with each other through local rules.  Even when rules and subsystems are simple, their interconnection can lead to a bewildering set of complex behaviors that scientists are struggling to understand. This talk attempts to understand such systems by taking an engineering approach: how should we build, design and control a system so that it behaves in a desirable, robust and predictable way? 

We first explain why, in general, designing and controlling large-scale distributed systems is computationally hard.  However, if a system is built to satisfy certain architectural properties, then these difficulties are reduced.  Different architectural choices impact the performance, robustness and complexity of a system, and recent research provides a unifying theory of architecture for complex systems that allows this impact to be quantified.  This theory leverages recent deep theoretical results integrating distributed control and optimization, including much novel work unique to our group.  Throughout the talk, we make liberal use of case studies drawn from a wide breadth of application areas, as well as interactive live-demonstrations, to illustrate these theoretical concepts in an accessible, relevant and hopefully entertaining way.