Meeting 52, 2023-10-20
- Bengt Sennblad
When is the meeting? | 20-Oktober-2023 at 10.00 - 11.00 |
Where is the meeting? | Zoom video-conference: http://meet.nbis.se/aiaio |
What is the meeting about? | From Graph Neural Networks to Transformers |
Who is the responsible presenter? | Erik Ylipää, AIDA data hub, NBIS, LiU |
Abstract:Â For a long time, one of the standard ways of performing machine learning was to transform all observations into fixed-dimensional features before feeding them into some learnable model. While methods to handle differently sized inputs have been around (e.g. specialized kernels or Recurrent Neural Networks), these have mostly been applied in niche domains. With the emergence of Geometric Deep Learning, we have tools which are now very capable of handling variable sized inputs. This presentation will be an introduction to Graph Neural Networks (GNNs) and how they can be used to model large varieties of problems. We will have a look at a common structure of GNNs and how we can extend them into Transformer neural networks.
Slides: Can be found here