Erol Gelenbe is currently the “Dennis Gabor” professor in Electrical and Electronic Engineering at Imperial College. He received a PhD on “Stochastic Automata with Structural Restrictions” from the Polytechnic Institute of New York (NYU), and a Doctor of Science degree on “Modeles de Performances de Systemes Informatiques” from University Pierre et Marie Curie (Paris). A Fellow of ACM and IEEE and of several National Academies, his honours include the Commendatore al Merito of Italy and France’s Legion d’Honneur. He was awarded Doctorates Honoris Causa by the University of Rome Tor-Vergata, the University of Liege (Belgium) and Bogazici Universty (Istanbul). 

Abstract: The biological sciences have long been a source of inspiration in the design of solutions for complex problems posed by operations research. Neuronal model-based optimization, prey-predator models and food webs, ant colony optimization, reinforcement learning, data classification with neural networks, learning-based techniques for tracking and control, are some widely used techniques in operations research that have been inspired by nature. This lecture will delve deeper into this interface. I will first discuss spiking random neural networks and show that the Random Neural Network (RNN) is a rigorous mathematical model with function approximation capability and a remarkable product form solution, whereby in equilibrium the joint probability distribution of an arbitrarily large set of fully connected spiking neurons have a joint probability distribution which is the product of the marginal distributions of individual neurons. This initial discovery provided a rigorous basis for fast gradient and deep learning, which has been exploited in numerous applications,  including the design of a Cognitive Packet Network that is controlled via neuronal distributed intelligence to enhance Quality of Service and Security. The RNN has also been be used to provide more “truthful” information from web searches. The RNN has been generalized, giving rise to a branch of Queueing Theory called G-Networks which allow the rigorous prediction of the performance of distributed computer systems and data networks which incorporate dynamic control functions such as state dependent rerouting, load balancing and the elimination of overload, the reset of a system after failures, and the ability to modify a sub-system’s state with knowledge from other sub-systems. In this talk, in addition to the RNN and G-Networks, I shall also discuss how such models can be applied to the analysis of Gene Regulatory Networks and the detection of anomalies from micro-array data.

The Colloquium will be livestreamed from the ICTP website. Light refreshments will be served after the event. All are welcome.