Projects
A selection of projects that I'm not too ashamed of
Towards Strict Conservation of Quadratic Conservation Laws in Operator Learning (MSc Thesis)
In this project, I improved the performance of several operator learning frameworks by enforcing energy conservation. My work was supervised by Georg Maierhofer and Nicolas Boulé at Oxford/Imperial.
Discrete Morse Theory for Computing Mayer-Vietoris Long Exact Sequences for Cellular Cosheaf Homology (Course Paper)
In this project, I explain computational aspects of computing Mayer-Vietoris LES for the case of Cosheaf Homology.
sPINNacle: Scale-Invariant Collocation Point Selection with NTK Learning Rate Adaptation (Course Paper)
In this project, I combine automatic collocation point selection with an NTK-based learning rate adaptation scheme to improve the robustness of PINNs under domain rescaling.
Discrete Flow Sampling with Trajectory Balance (Research Proposal)
This proposal is based on insights from my research project on "Training Energy-Based Models by Learning to Sample with Discrete Flows."
Training Energy-Based Models by Learning to Sample with Discrete Flows (MSc Thesis)
In this research project I used discrete flows to amortize the generation of negative samples for training energy-based models. My work was supervised by Yingzhen Li and Zijing Ou at Imperial.
Graph-based Super-Resolution of Brain Connectivity Matrices (Coursework)
In this group project we developed a new framework for brain graph super-resolution.
Detecting Patronizing Language (Coursework)
In this project I fine-tuned a RoBERTa LLM for the task of detecting partonizing and condescending language.
Sovereign Commitment in ECB Announcements (BSc Thesis)
In this research project I conducted a factor analysis of the market reaction to ECB announcements, and was able to identify a novel causal channel that I label "Sovereign Commitment". This work was supervised by Shengxing Zhang at LSE.
