ODSL Forum: Optimization and interpretability of graph attention networks for small sparse graph structures in automotive applications

by Marion Neumeier (TH Ingolstadt, CARISSMA Institute of Automated Driving)

Friday 11 Aug 2023, 14:00 → 15:00 Europe/Berlin

Basement Seminar Room (Origins Cluster Building)

We are delighted to invite you to our next ODSL Forum with a talk from industry!
Marion Neumeier is pursuing her PhD at the TH Ingolstadt in a collaboration with the CARISSMA Institute of Automated Driving and she will introduce us to her work on graph neural networks for automotive applications!

We cordially invite you to attend his talk on Friday, August 11th at 2 pm.
Marion will come to Munich for the talk and give it in the Origins basement seminar room. As usual, we will set up a live-stream via Zoom for anyone that cannot attend in person.

Title: Optimization and interpretability of graph attention networks for small sparse graph structures in automotive applications
Abstract: Graph Attention Networks (GATs) have proven to be powerful frameworks for processing graph-structured data and, hence, have been used in a range of applications. GATs tackle the problem of high-dimensional learning by introducing geometric priors, treating the input data as graphs and applying attention mechanism to perform weighted message-passing. However, achieved performance by these attempts has been found to be inconsistent across different datasets and the reasons for this remains an open research question. Building on the fundamentals of GATs, we will discuss potential pitfalls of GATs that hinder an optimal parameter learning and interpretability of attention scores as relevance indicator. The general discussion will be accompanied by application examples and evaluations from the automotive field.

The Zoom Link is available on request.

2023_08_11_OptGraphStruct.pdf