Biography

I was first introduced to the enormous potential of statistical modelling and machine learning during my undergraduate degree at the LSE and master’s degree at Imperial College. Later on, I worked as a Data Scientist to optimize API production yield in large-scale production for biotechnology and pharmaceutical companies, where I noticed a large gap between assumptions made in statistical theory and real-world applications. While deep learning models have the potential to improve many applications in biotechnology research and medicine, their reliance on vast, thoroughly labeled datasets often presents a significant bottleneck to real-world adoption. Additionally, their lack of transparency impedes the trust by practitioners, researchers, and patients. I have explored this problem throughout my Computer Science MPhil at Cambridge by studying how we can use design more intuitive decision-making tools for different cancer screening tasks. During my PhD, I am developing multimodal fusion and self-supervised learning methods for heterogeneous biomedical data modalities. My research focuses on learning meaningful representations of human tissue using histology, bulk sequencing, and spatial transcriptomics data from cancer biopsies without relying on exhaustive manual annotation to improve common diagnostic tasks like tumor staging, subtyping, and predicting treatment outcomes.