Swedish web site
Graph Learning2024VTFull
|
|
Course plan
No of lectures
Depending on the number of participants, there will be around 10 meetings.
Recommended for
PhD students who want to get a better understanding of different neural network architectures for graph inputs.
Goals
The aim of the course is to get a good understanding of the available neural network architectures that are suited of graph inputs. We want to understand the expressive power and limitations of different architectures and see how such networks can be applied in the context of symbolic reasoning methods.
Prerequisites
A solid background in computer science, machine learning, and complexity theory is beneficial.
Organization
The course is run as a weekly seminar in which one participant presents a paper and leads a following discussion on the topic.
Content
In this seminar, we cover different neural-network architectures for learning
using graph representations. This includes, but is not limited to, the
following topics:
Graph Convolutional Networks (GCN)
Graph Neural Networks (GNN)
Message-Passing Neural Networks (MPNN)
Graph Transformer Architectures
Expressive Power of different Architectures
Applications in Symbolic Reasoning
Literature
A selection of recent papers will be provided on the course webpage.
Examination
Presentation of a research article, possibly including some coding examples, leading the discussion on the presented topic, active participation in seminar discussions.
Examiner
Daniel Gnad
Credits
6
Page responsible: Director of Graduate Studies
