RESEARCH DIRECTION / 02
Deep Learning
Learning representations that support geometric reasoning.
Conceptual illustration: a feature field on a geometric manifold. Not a trained model or an experimental result.
Learning with geometry in mind
Deep learning is one of my core research interests, alongside geometric perception and uncertainty quantification. I am interested in how learned representations can support perception systems that work across different kinds of sensor observations.
The broader motivation is to connect flexible, data-driven representations with the geometric structure of the physical world.
Questions I want to explore
- How can learned representations retain information that matters for 3D geometric reasoning?
- Which properties of a representation can transfer across sensing modalities?
- How can learning and geometric estimation complement each other when observations are imperfect?
Part of a larger perception system
Learning is closely connected to both geometric perception and uncertainty quantification. My interest is in their intersection: representations that are useful for geometry, and estimates whose reliability can be understood.