Skip to main content
Harry Anthony, DPhil Student


Harry Anthony MSc

DPhil Student


Harry Anthony is a DPhil student at the University of Oxford, under the supervision of Professor Konstantinos Kamnitsas. As part of the IBME group, Harry’s research focuses on improving the reliability of deep neural networks in the field of medical imaging, with a particular emphasis on out-of-distribution (OOD) detection. OOD detection is critical for ensuring the accuracy and safety of neural networks for medical imaging analysis, as it helps to identify when input data differs from its training data. Harry’s research aims to develop novel methods for deep learning algorithms to detect OOD inputs, as well as improving the overall performance of medical imaging analysis. Prior to his doctoral studies, Harry obtained a first class master’s degree in Physics from Imperial College London.

Awards and Prizes

Best Paper Award at the UNSURE workshop at MICCAI 2023

Research Interests

Harry’s primary research interests revolve around out-of-distribution (OOD) detection and uncertainty quantification for medical image analysis using deep neural networks. Given that neural networks may produce unreliable and erroneous predictions when confronted with inputs that differ significantly from the training data (OOD), the ability to identify and mitigate such instances is important. This is a significant issue for AI in medical image analysis, as wrong predictions on OOD inputs could have serious implications for decisions made downstream.

Harry’s research encompasses a variety of approaches to out-of-distribution detection. This includes developing OOD detectors tailored for pre-trained models, as well as developing new model architectures and training techniques specifically designed for OOD detection. The key aim of this research is to enable models to discern when they are presented with unfamiliar data, hence being able to detect when they may give an erroneous prediction (known as “failing gracefully”). Developing AI models with these abilities is an important step towards improving the robustness and reliability of deep neural networks for medical image analysis.

Research Groups


Related Academics