Bio
Professor Adam Sobey is Programme Director of Data-Centric Engineering at The Alan Turing Institute and Professor of Data-Centric Engineering in the Maritime Engineering group at the University of Southampton. He is also a Non-Executive Director at Theyr Ltd, a MetOcean and Voyage Optimisation Software provider, and at AQ, provide tools for Digital Assurance.
He completed his degree in Astronautics at the University of Southampton in 2006 and his PhD in the Maritime Engineering group in 2010 using AI in the design of leisure boats. From 2009 he managed the Lloyd’s Register/Ministry of Defence Centre of Excellence in Marine Structures at Southampton, developing new techniques to model damaged ships. This research was incorporated into Lloyd’s Register’s design guidance and in 2015 he was awarded the Royal Institution of Naval Architect’s Jeom Paik Award for stochastic analysis of ship structures. In 2014 he became a Lloyd’s Register Educational Trust funded Research Fellow with 50% of his time seconded to the Institute of High Performance Computing in Singapore, developing novel algorithms in Evolutionary Computation. In 2015 he was a co-Investigator for the University of Southampton Lloyd’s Register Foundation University Technology Centre in “Ship Design for Enhanced Environmental Performance”. He became a lecturer in the Maritime Engineering group in 2018 and was asked to start a theme in Marine and Maritime within the Data-Centric Engineering programme at The Alan Turing Institute in 2019 with Gabriel Weymouth. He became the first Professor in Data-Centric Engineering in 2022 and Programme Director for Data-Centric Engineering in 2023.
He is an Executive Editor for the Data-Centric Engineering journal and Ships and Offshore Structures. He has generated a range of products from his research, mainly focused on decarbonisation in Shipping.
Research interests
Adam is interested in Reinforcement Learning, Machine Learning and Evolutionary Computation approaches. In particular:
- co-evolutionary Multi-Level Selection Genetic Algorithm (cMLSGA) inspired by multi-level selection theory and in using Epigenetic algorithms for dynamic search problems.
- An approach to machine learning that fits to the ground truth when we don’t know what the ground truth is, allowing stronger generalisation off dataset.
- The potential for multi-policy learning that allows improved transfer learning with reduced data and memory. This is based on the idea of Task Capacity, that a given policy will have a maximum number of tasks that it can learn effectively depending on the interference of the tasks and the size of the policy.
