Toward Data-Efficient Reinforcement Learning

We are surrounded by large-scale interconnected systems, from the Internet to the power grid, from the commuter traffic flow to social networks. While essential, the management of such networked systems is exceedingly hard, mainly because of their intrinsic and constantly growing complexity. To overcome this challenge, we are in dire need of data-efficient online decision strategies under uncertainty for high-dimensional and dynamic networks. This project focuses on developing fundamental learning methodologies targeting data-efficiency in large-scale problems, specifically, data-efficient reinforcement learning methodologies able to infer and exploit the structure of the network. . This project will develop fundamental methodologies on data-efficient AI and autonomous systems, aligning with the S&T Challenge of AI and the "AI technologies", and "decision making strategy" EPSRC research areas - some of the largest/growing portfolios. This research directly links also to UKRI grand challenge on AI, the EPSRC areas as productivity and connectedness, developing "cost-effective ways of delivering existing services" in a fully connected world [EPSRC Delivery Plan, 2016-2019], in the envisioned application of controlling IoT networks. It links indirectly also the following UCL Grand Challenges: the UCL Grand Challenge of Sustainable Cities (AI for autonomous driving cars, grid energy, energy-harvesting IoT networks); the UCL Grand Challenge of Transformative Technology (AI for managing Internet of Food, understanding social behaviours/impact); the UCL Grand Challenge of Cultural Understanding (graph learning to influence of religions, politics opinion across the world, AI for personalized education).