Edward Phillips

I am a DPhil candidate in Engineering Science at the University of Oxford, working in the Computational Health Informatics Lab under Professor David Clifton.

My research develops reliable intelligent systems for clinical decision-making. My past and current work includes:

Uncertainty quantification in large language models. Methods for detecting and mitigating hallucinations, including Semantic Self-Distillation (UAI 2026) and geometric approaches to LLM uncertainty (TMLR 2026).
Guideline-grounded treatment planning agents. An agent-based framework for treatment planning in oncology, with an upcoming pilot evaluation in collaboration with NHS and industry partners (TrustedMDT).
Early disease diagnosis from large-scale hospital records. Foundation models for electronic health record data, currently in development as part of CODETECT, a multi-site NHS study identifying hospitalised patients with undiagnosed chronic conditions.

Before my DPhil I spent three years in industry as a data scientist and systems engineer, working in generative AI, medical device development, and biomarker research. Most recently I completed a research internship at Spotify, developing efficient and uncertainty-aware LLM ranking systems. I hold a Master’s Degree in Information and Computer Engineering from the University of Cambridge, achieving First Class with Distinction.

selected publications

UAI
Semantic Self-Distillation for Language Model Uncertainty

Edward Phillips, Sean Wu, Fredrik K. Gustafsson, and 2 more authors

Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence, 17–21 Aug 2026

Abs arXiv Bib PMLR PDF Poster

Large language models present challenges for principled uncertainty quantification, in part due to their complexity and the diversity of their outputs. Semantic dispersion, or the variance in the meaning of sampled answers, has been proposed as a useful proxy for model uncertainty, but the associated computational cost prohibits its use in latency-critical applications. We show that sampled semantic distributions can be distilled into lightweight student models which estimate a prompt-conditioned density before the language model generates an answer token. The student model predicts a semantic distribution over possible answers; the entropy of this distribution provides a prompt-level uncertainty signal, and the probability density allows answer-level reliability evaluation. Across experiments on TriviaQA and MMLU, we find our student models perform competitively relative to the teacher’s sampled semantic dispersion on a hallucination prediction task, whilst offering additional uncertainty primitives for out-of-domain detection and multiple-choice answer selection. We term this technique Semantic Self-Distillation (SSD), which can serve as a general framework for distilling predictive uncertainty in complex output spaces beyond language.
@inproceedings{phillips2026ssd, title = {Semantic Self-Distillation for Language Model Uncertainty}, author = {Phillips, Edward and Wu, Sean and Gustafsson, Fredrik K. and Gao, Boyan and Clifton, David A.}, booktitle = {Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence}, editor = {Perković, Emilija and Malinsky, Daniel}, volume = {337}, series = {Proceedings of Machine Learning Research}, pages = {5427--5447}, month = {17--21 Aug}, publisher = {PMLR}, year = {2026}, }

TMLR

Geometric Uncertainty for Detecting and Correcting Hallucinations in LLMs

Edward Phillips, Sean Wu, Soheila Molaei, and 3 more authors

Transactions on Machine Learning Research, 2026

arXiv Bib OpenReview

@article{phillips2026geometric,
  title = {Geometric Uncertainty for Detecting and Correcting Hallucinations in {LLM}s},
  author = {Phillips, Edward and Wu, Sean and Molaei, Soheila and Belgrave, Danielle and Thakur, Anshul and Clifton, David A.},
  journal = {Transactions on Machine Learning Research},
  issn = {2835-8856},
  year = {2026},
}