PhD in Trustworthy and Efficient AI Systems

About the project

About the Project About the Project This PhD will focus on developing trustworthy and efficient AI systems built around large language models (LLMs) and related foundation models, covering topics such as reliability and truthfulness, efficient and frugal LLM inference and deployment, multimodal learning and reasoning, privacy-preserving and federated training, and trustworthy multi-agent and collaborative AI systems, with applications across domains such as healthcare, communication networks, and other time-critical settings. You will explore methods that ensure these systems are reliable, adaptive, privacy-preserving, and resource-efficient in real-world applications. Key Research Themes (indicative, not exhaustive) • Reliable and Truthful LLMs/VLMs: Ensuring large language and vision-language models are robust and trustworthy. • Efficient LLM Inference and Deployment: Techniques such as model compression, distillation, routing, and latency-aware optimisation to enable fast and frugal LLM use at scale and at the edge. • Multimodal Learning and Reasoning: Methods that combine text, images, and other modalities (e.g., audio, structured data) for more grounded and context-aware AI systems. • Multi-Agent AI Systems: Coordination, routing, and learning among multiple AI agents in complex, heterogeneous settings. • Privacy-Preserving Training/Fine-Tuning of Foundation Models: Protecting user data (e.g. via federated learning, differential privacy, secure aggregation) while training or adapting large models. • Trustworthy Federated/Collaborative Learning: Ensuring integrity, verifiability, robustness, and security in collaborative training settings. You will be encouraged to shape the exact topic based on your interests within these themes. Candidate Profile Prospective candidates should have: • A Bachelor’s degree with a minimum of 2:1 (or international equivalent); a Master’s degree is desirable. • Strong foundation in computer science, AI, or a closely related field. • Proficiency in mathematics, including linear algebra, probability and statistics, multivariable calculus, and optimisation. • Proficiency in programming (e.g., Python), with proven ability to implement neural networks, transformers, encoder-decoder models, and other core components of LLMs. • Familiarity with, or keen interest in, trustworthy AI and/or efficient LLMs and foundation models. • Excellent oral and written communication skills. • Commitment to high-quality, rigorous research. Environment The University of Birmingham ranks among the top 100 universities worldwide and is a member of the Russell Group of UK research-intensive universities. The School of Computer Science is ranked 3rd in the UK for world-leading research (REF 2021). The School’s research spans AI, cybersecurity, computational theory, and socio-technical systems. This PhD will cut across several of these areas, offering opportunities for interdisciplinary collaboration. We particularly welcome applications from ethnic minorities, underrepresented groups, individuals with disabilities, and neurodiverse candidates. How to Apply For guaranteed full consideration, apply by 4 December 2025, 5 PM UK time via this link and select Dr. Baturalp Buyukates as your supervisor. If you are interested, please also send an email to Dr. Baturalp Buyukates at b.buyukates@bham.ac.uk including: • Your CV • Academic transcripts • One representative research paper (if applicable) Shortlisted applicants will be invited to: 1. Give a short presentation on one or more of the PhD themes, and 2. Complete a brief technical exercise.

Research areas