👋 Hi, I'm Jonathan Rubin, a PhD candidate at Imperial College London. I am passionate about building statistical foundations to novel machine learning paradigms, for better explainability and usage in scientific discovery. My research ultimately aims to improve the prognosis and treatment of metastatic cancer by developing and leveraging tools in Variational Inference, Phylogenetic Inference, and Geometric Machine Learning.

I graduated with a MSci in Mathematics, First Class Honours, from Imperial College London in 2023, where I gained extensive experience in Machine Learning, Computer Vision, and Geometric Deep Learning. I have also worked as a Machine Learning intern at OpenOcean, a European AI focused Venture Capital firm, where I implemented a full ML pipeline on Microsoft Azure, using graph neural networks to predict the likelihood of start-ups raising funding rounds. Additionally, I have participated in interdisciplinary research projects in computer vision with the Medicine Department at Imperial, and in data science contests with Citadel.

📚 Education:

🔹 Imperial College London - PhD: AI & Machine Learning @ AI4Health CDT

🔹 Imperial College London - MSci: Mathematics (First Class Honours)

🔹 King Edward VII School - A Levels: Mathematics, Physics, Further Mathematics, Chemistry

📖 Master's Dissertation Thesis (Paper coming soon):

🔹 Title: "A Statistical Geodesic Perspective on Heterophilic Bottlenecking in Graph Neural Networks"

🔹 Supervisors: Prof. N. Jones and Dr. S. Loomba

🔹 In this research, I explore the impact of homophily and heterophily on the performance of Graph Neural Networks (GNNs). The study focuses on understanding the phenomena of bottlenecking and underreaching in SBM model random graphs.

🔹Upcoming conference paper titled: "Geodesic Distributions Reveal How Heterophily and Bottlenecks Limit the Expressive Power of Message Passing Neural Networks" - In this paper we expand on many of the notions developed in my thesis, to develop a statistical, Jacobian based theoretical foundation for feature expressivity and generalisation in machine learning classification tasks, allowing a bottom up approach for understanding the combined effects of bottlenecking and heterophily on GNN classification perforamnce

🏆 Competitions & Achievements:

🔹 Shortlisted for Citadel Europe Datathon (2021)

🔹 Imperial College Integration Bee (2021) – 3rd Place

🔹 National Cypher Challenge (2018) - 7th place (out of 400)

🔧 Skills:

🔹 Machine Learning

🔹 Data Science

🔹 Natural Language Processing

🔹 Computer Vision

🔹 Graph Neural Networks

🔹 R and Python programming languages

🔬 Research Experience:

🔹 Upcoming conference paper: "Geodesic Distributions Reveal How Heterophily and Bottlenecks Limit the Expressive Power of Message Passing Neural Networks"

🔹 Computer Vision Research Project: "Deep Learning Based Airway Segmentation for High Resolution CT Images to Facilitate COVID-19 and Lung Fibrosis Diagnosis and Prognosis"

🎸 Interests:

🔹 Drumming

🔹 Writing

🌐 Connect with me:

🔹 LinkedIn: Jonathan Rubin

🔹 Email: jonathan.rubin19@imperial.ac.uk

Feel free to explore my repositories and don't hesitate to reach out if you'd like to collaborate on a project or discuss ideas!