Analysis of complex heterogeneous networks: scalability, robustness and fundamental limitations

I. C. Lestas

About the Event

Complex networks are receiving an increasing attention by various scientific communities, as a result of their sig- nificance and enormous impact in both natural and man-made systems. Such examples could range from Internet protocols, power networks, vehicle platoons to flocking phenomena and gene regulatory networks. On the one hand it is important to understand the fundamental principles and theories behind the success of networks present in nature, but, at the same time, there is an urgent need to develop methodologies that enable the design of networks where robustness and scalability can be guaranteed. This talk is going to address such challenges by presenting recently developed tools that can lead to scalable network designs. Furthermore it will be shown how combining control and information theory can lead to powerful tools for the analysis of biological networks and the derivation of fundamental limitations in their performance. By scalability we refer to the requirement that robust stability is guaranteed for an arbitrary interconnection by conditions on only local interactions, without having to redesign the whole network whenever a new heterogeneous agent is added or removed. It would be, for example, unrealistic to carry out a centralized analysis, whenever a computer/router enters the Internet or a generator becomes part of a power network. We show in the talk, how appropriate convexification arguments can lead to stability certificates that are both decentralized and scalable. This creates an abstraction that is relevant in diverse applications such as data network protocols and group coordination problems. For the case of biological networks we focus on life processes at the molecular level. These are well known to be inherently stochastic, with a substantial part of the noise being intrinsic, arising from the random births and deaths of individual molecules. Such fluctuations are crucial in the functionality of many biological processes, as noise ran- domizes signaling pathways and drives metabolites away from optimal concentrations. Understanding these effects is currently a major challenge due to the complexity of the underlying biochemical reaction networks which are often very poorly characterized. It will be shown in the talk how combining control and information theory can lead to fun- damental limits for the suppression of noise in gene regulatory networks. These are hard bounds that hold for arbitrary feedback, as a result of features that restrict the ability to transmit information reliably. These results have important biological implications as they have provided an explanation to some striking experimental observations where a huge amount of resources are used within the cell for noise suppression. They also show how an interface between control and information theory can be used to rigorously analyze poorly characterized biological systems.

Biography

Ioannis Lestas received the B.A. (Starred First) and M.Eng. (Distinction) degrees in Electrical Engineering and Infor- mation Sciences and a Ph.D. in control engineering from the University of Cambridge (Trinity College). His doctoral work was performed as a Gates Scholar and Trinity College Research Scholar. In 2006 he was elected to a Junior Research Fellow of Clare College, University of Cambridge; he was awarded a five year Royal Academy of Engineer- ing research fellowship in 2008, and since October 2009 he is an official Fellow of Clare College and a Director of Studies in Engineering at Cambridge. His research interests lie within the area of analysis and control of large scale networks, focusing on the development of frameworks for addressing issues of scalability, robustness and fundamental limitations, with applications in data networks, multiagent systems and biological networks.

Additional Information

Contact: Beth Olsen

Phone: 734 763-8040

Email: bethi@umich.edu

Sponsor: ECE

Open to: Public