PhD Studentship - Protein Quantum Logic Gates

Location: Streatham Campus

About the Project Project details

Self-adaptive and autonomous systems are increasingly deployed in dynamic and uncertain environments, where effective decision-making relies on accurate models of how

the system interacts with its environment. Traditionally, such world models—comprising transition and observation models—are specified by domain experts, based on their knowledge and assumptions about the system and its operational context. However, as environmental conditions evolve, these assumptions may no longer hold, leading to inaccurate predictions of adaptation impacts and suboptimal decision-making. Examples include crowd management and large-scale communication networks based on cellular or wireless sensors. For instance, during mass gatherings such as the sport matches (e.g., Superbowl) or large music festivals, assumptions of steady cellular network demand often fail. Sudden surges in GSM/5G traffic, stampede risks, or emergency events introduce uncertainty that challenges existing models of reliability, performance, and safety. Similarly, in dynamic crowd scenarios, assumptions about orderly movement can break down due to panic or unexpected human behaviour, leading to dangerous scenarios like stampedes. Managing such environments requires robust, real-time adaptation that accounts for evolving conditions and uncertainties. More application examples include real world environments like health care and environmental monitoring.

This PhD project aims to address these challenges by exploring how evolutionary algorithms and reinforcement learning (RL) techniques can be combined to continuously learn, adapt, and refine world models in self-adaptive and autonomous systems. Specifically, the research will investigate how AI-based methods can support the evolution and updating of transition and observation models to reflect real-time changes in environmental conditions, enabling more accurate predictions of adaptation impacts and thereby supporting a better-informed, resilient decision-making process.

Research Objectives Model Learning in Dynamic Contexts Investigate the use of reinforcement learning for constructing and updating probabilistic world models (transition and observation functions). Explore model-based RL approaches that integrate learned models with planning and adaptation mechanisms. Hybrid Evolutionary-RL Framework Develop novel frameworks with evolutionary algorithms to search optimised model structures, complementing RL’s incremental updates. Consideration of robustness to various sources would be an objective too.

Expected Contributions A novel framework combining evolutionary algorithms and reinforcement learning for adaptive world model learning. Methods for continuously refining expert-specified models based on observed environmental feedback. Decision-making approaches that integrate updated models to improve adaptation strategies in uncertain conditions. Demonstration of applicability through experimental validation in real-world inspired scenarios.

Impact This research will advance the foundations of adaptive and autonomous systems by enabling them to maintain accurate and up-to-date models of their environments. The ability to learn world models dynamically will significantly enhance their capacity for self-management, leading to more reliable, efficient, and trustworthy systems in safety-critical and complex domains.

On-the job research training for the Doctoral Researcher. The PhD researcher will gain expertise in implementing RL and Evolutionary Optimisation to a real-world usecase on dynamic crowd and network management provided by the industrial partner INOCESS Inc., France (https://www.inocess.com/?page_id=1370&lang=en). The industrial partner will provide co-supervision, access to relevant data, offer opportunities for field testing of the proposed solutions, and may host the PhD candidate at its facilities.

Please direct project specific enquiries to: Dr Huma Samin (h.samin@exeter.ac.uk)

Please ensure you read the entry requirements for the potential programme you are applying for. To Apply for this project please click on the ‘Apply’ button above.

Advert information