Choose timezone

Your profile timezone:

Use timezone based on:

Event/category Custom

Select a custom timezone

Login

IPA Colloquium FS25

Thursday 3 Apr 2025, 12:00 → 13:00 Europe/Zurich

32/1-A24 (CERN)

32/1-A24

CERN

40

Show room on map

- 12:05 → 13:00
  Talk 32/S-C22
  
  32/S-C22
  
  CERN
  
  17
  Show room on map
  - 12:05
    
    Efficient architectures with geometric deep learning 45m
    
    Geometric deep learning leverages symmetry and transformation properties of input and output spaces to design neural network architectures. By incorporating invariance or equivariance into the model, the resulting architecture can be easier to learn, achieve better generalization performance, and use fewer parameters. Recent equivariant architectures show competitive compute requirements, but there are also open questions on the trade-off between exact symmetries and soft inductive bias towards simpler models.
    
    Speaker: Dr Hampus Linander (University Göthenburg)