Bio
Simon is a Ph.D. candidate in AI Enabled Healthcare at the University College London’s Centre of Doctoral Training based in the Institute of Health Informatics. Previously, he studied for an MSc in Computer Science at Imperial College London and a BSC in Natural Sciences at Durham University. Simon has also spent time in industry, including a year-long placement in a clinical research organisation’s (Parexel) artificial intelligence lab in San Francisco as part of the Silicon Valley internship programme.
Research interests
Recent work has shown that the principles of large-scale next-token prediction (or autoregressive) training, used to train models such as GPT-4, can be adapted to the healthcare domain. Instead of predicting the next word in a sentence, the training objective is altered to predict the next event (Snomed-CT code) in a patient’s timeline. The efficacy of this approach was shown using data from a single hospital trust. The next stage of Simon’s PhD proposes to scale this method to a nationwide dataset of 60 million patient records. We hypothesise that the benefits of scaling seen in large language models could also occur in healthcare. So, by coupling the latest advancements in generative AI with this unique dataset, we aim to build the first national-scale healthcare AI model. Analogous to the apparent emergent behaviour of language models, the potential use cases for such a model are a rich area of exploration. Current hypotheses include virtual screening, identifying individuals at high risk of disease or emergency admission and simulating clinical trials to predict the effect of changing medications.