Events

Date: 7 October 2025   Time: 15:00 - 16:00

Location: [Room] ME Eng 101a

Speaker: Dr. Alec Diallo

Moderator: Dr Ignacio De Castro Arribas

Abstract: More data has long been the recipe for better machine learning. But what if the real frontier is not in scale, but in structure? In this talk, I will show how hidden geometries of data, captured through Proximally-Connected graphs, can be made explicit and actionable. This structural view allows us to determine when meaningful patterns exist, and to distill large datasets into compact subsets that retain their full representational and decision-making power. Specifically, I will show how structural reasoning can set new standards for data analysis and data-efficient learning. And more broadly, I will argue that moving from a scale-centric to a structure-centric view of data reshapes the very foundations of learning: one where models can learn from fewer resources and generalize more faithfully to the patterns that truly define knowledge.

Bio: Alec Diallo is a Postdoctoral Research Associate in the Mobile Intelligence Lab and a member of the Institute for Computing Systems Architecture at the University of Edinburgh. He earned his PhD in Informatics at Edinburgh, funded by Arm, where his research on AI and cybersecurity was recognized with the Brendan Murphy Memorial Prize and the SICSA Best PhD Dissertation awards. His current research focuses on computational intelligence methods and their applications, particularly on knowledge representation, optimization strategies for learning algorithms, and the security and privacy of AI systems.