Trustworthy Autonomous AI Agents via Adaptive Model Selection and Formal Uncertainty Quantification The University of Manchester in United Kingdom

This PhD opportunity at The University of Manchester focuses on developing trustworthy, uncertainty-aware autonomous AI agents capable of robust operation in dynamic, partially-observable environments. The project addresses key challenges in deploying autonomous systems, such as distribution shift, missing information, and unmodelled factors, which can lead to fragile behaviour and silent failures. Current solutions often trade off reliability for computational efficiency, highlighting the need for new approaches that balance safety and performance.

The research will integrate formal uncertainty quantification, adaptive model selection, and sequential decision-making under non-stationarity. Techniques such as conformal prediction for distribution-free uncertainty, non-stationary multi-armed bandits for adaptive control, and expert model aggregation will be explored. The goal is to enable agents to dynamically select among specialised predictive models while maintaining provable performance guarantees, advancing both theoretical understanding and practical implementation.

The successful candidate will contribute to open-source software and theoretical advances, with applications in high-stakes domains including autonomous systems, healthcare, and robotics. The project is based in the Department of Computer Science and supervised by Dr M Hosseini.

Eligibility: Applicants should hold or expect to achieve at least a 2.1 honours degree or a master’s (or international equivalent) in a relevant science or engineering discipline. Preferred backgrounds include computer science, AI, formal methods, cybersecurity, cryptography, or mathematics. Outstanding candidates with a four-year bachelor's degree are also encouraged to apply. While publication is not mandatory, a high-quality publication can strengthen applications, especially for overseas candidates seeking funding.

Funding: Excellent candidates will be nominated for competence-based faculty funding, which covers tuition fees and provides a tax-free stipend at the UKRI rate (£20,780 for 2025/26), with expected annual increases. Self-funded students are welcome to apply. The start date is October 2026.

Application Process: Apply online via the university’s portal. Early application is recommended as the advert may be removed before the deadline. Applicants must submit all required documents, including transcripts, CV, supporting statement, and referee contact details. English language certification is required if applicable. Contact the supervisor before applying to discuss your background and motivation for the project.

Equality, Diversity, and Inclusion: The University of Manchester is committed to diversity and encourages applicants from all backgrounds and career paths. Flexible study arrangements, including part-time options, may be available depending on the project and funding.

For further details, visit the project page or contact the admissions team at [email protected].