PhD Researcher at the intersection of non-Euclidean geometry and AI
Hyperbolic Learning โข Concept Representation โข Machine Unlearning
I'm investigating how non-Euclidean geometries can revolutionize AI's ability to understand hierarchical concepts and selectively forget information. My work on hyperbolic contrastive unlearning explores the fundamental question:
"How does the underlying geometry of representation space affect our ability to edit and remove concepts from AI models?"
๐ Research Foundation
- ๐ญ PhD Researcher at Visual Perception Analysis Lab, Aalborg University
- ๐งโ๐ซ Supervised by Prof. Thomas B. Moeslund, Assoc. Prof. Kamal Nasrollahi, and Prof. Sergio Escalera
- ๐ฐ Research funded by the Milestone Research Programme
- ๐ Splitting time between Barcelona & Aalborg
- ๐ MERU-Unlearn: Adapting unlearning mechanisms to hyperbolic representation spaces
- ๐ง Hierarchical Concept Learning: Understanding how concept hierarchies impact representation learning
- ๐ Geometric Contrastive Learning: Exploring the benefits of non-Euclidean spaces for contrastive representation learning