About
I'm Finn Sherry. I'm a PhD candidate in Remco Duits' Geometric Learning & Differential Geometry group at TU/e.
I am a part of the EAISI-funded
research project "New Memory Efficient Geometric Learning for Automatic Crack Detection in Bridges".
Our goal is to develop and improve Geometric Deep Learning (GDL) techniques.
Current GDL algorithms, such as Group-equivariant Convolutional Neural Networks (G-CNNs), have a proven track record in low data applications,
but often struggle when dealing with many large images.
We are investigating various ways to make scaleable GDL algorithms, e.g. by reducing their memory footprint.
In the end, we hope to apply our new techniques to automate the detection of cracks in infrastructure:
the vast quantity of high resolution images in this application makes current GDL methods impractical.
In addition to machine learning and (differential) geometry, I like analysis, PDE theory, and mathematical physics. I am also interested in history, particularly classical antiquity, and I'm a member of the Dutch classics association NKV.
