Description
This NGT Whiteboard Session will focus on applying tools from theoretical physics, geometry and topology towards understanding the propagation of information through Neural Networks as well as understanding their key properties such as generalisability and robustness.
This session will be organised on the occasion of Prof. Tolga Birdal’s visit to CERN for the CERN STEAM Academy Seminar on Topological Deep Learning for the Next Generation of AI4Science.
Following his seminar, this dedicated NGT-WHISE session will provide an open space for further discussion and making connections to research directions at CERN.
Topological Deep Learning extends machine learning beyond graphs, enabling models to represent higher-order interactions, motifs, cycles, surfaces, and richer topological structures.
As with all NGT WHISE sessions, the format will be intentionally lightweight, with minimal slides and a strong focus on open discussion and whiteboard-style exchange.
Coffee and snacks will be provided.
Guiding question
How can we guide Neural Networks towards learning maximally informative representation for optimal data acquisition and insight?
Format
This is an informal discussion session aimed at exploring early-stage ideas, possible research directions, and cross-WP connections. Participants are encouraged to join actively, ask questions, and contribute perspectives from their own work.
Guest
Prof. Tolga Birdal
Imperial College London
Registration
Please register so that we can estimate the number of participants.