Universal laws and architecture in complex networked systems

Effective layered architectures such as the brain seamlessly integrate high level goal and decision making and planning with fast lower level sensing, reflex, and action and facilitate learning, adaptation, augmentation (tools), and teamwork, while maintaining internal homeostasis.  This is all despite the severe demands such actions can put on the whole body’s physiology, and despite being implemented in highly energy efficient hardware that has distributed, sparse, quantized, noisy, delayed, and saturating sensing, communications, computing, and actuation. Similar layering extends downward into the cellular level, out into ecological and social systems, and many aspects of this convergent evolution will increasingly dominate our most advanced technologies. Simple demos using audience’s brains can highlight universal laws and architectures and their relevance to future network technologies.

With this motivation, we’ll introduce a new unified theory of complex networks that integrates communications, control, and computation with applications to cyberphysical systems as well as neuroscience and biology.  Though based on completely different constraints arising from different environments, functions, and hardware, such systems face universal tradeoffs (laws) in dimensions such as efficiency, robustness, security, speed, flexibility, and evaluability. And successful systems share remarkable universals in architecture, including layering and localization, to effectively manage these tradeoffs.  The study and design of systems architectures have traditionally been the among the areas of engineering least guided by theory, and there is nothing remotely resembling a “science” of architecture.  Our aim is to fundamentally change this with new mathematics and applications, while still gaining valuable insights from both the art and evolution of successful architectures in engineering and biology.

The course is given within the LCCC focus period on Learning and Adaptation for Sensorimotor Control. Course responsible is Professor John Doyle from Caltech, USA. Teaching assistant is Yorie Nakahira from Caltech, USA.

The course will be taught at 10h30 - 15h00 on October 29-31 in the seminar room at the Department of Automatic Control (M:2112B).