Bio
Michael Doherty is a PhD student in the Optical Networks Group at University College London, and part of the Connected Electronic and Photonic Systems (CEPS) Centre for Doctoral Training (CDT). His research develops optimisation techniques to improve the data throughput, energy efficiency and quality of service of future fibre optic communication networks, with a particular focus on Reinforcement Learning (RL). His interests extend to a variety of graph-based combinatorial optimisation problems that occur in optical networks, and the potential of fast simulations combined with RL to find high-quality solutions. He completed a MSci in Physics at Imperial College London in 2017 and gained experience at the National Physical Laboratory and as a data engineer in industry before joining the CEPS CDT.
Research interests
Michael's research at the Turing will investigate the potential to reduce the carbon emissions of optical data communication networks through the exploitation of two factors: 1) The spatio-temporal variation in carbon intensity of energy sources. 2) The energy use characteristics of various network hardware (e.g. IP routers and optical transponders). Combining these two sources of information and translating that into network management decisions is non-trivial. By using model-based reinforcement learning (the same kind of “game-playing” AI used to master boardgames and other tasks), Michael will develop models capable of managing network traffic while minimizing carbon emissions, subject to quality-of-service constraints. The goal then will be to ascertain how helpful these techniques are in reducing the environmental impact of this digital infrastructure compared to other methods. As these networks constitute vital national infrastructure, the issues of reliability, explainable decisions and security also need to be addressed.