Generating minimal timeless traffic-like datasets as a basis for applying geometric deep learning to congestion prediction

Digital seminar with phd-candidate Grunde Wesenberg December 20th at 10.00-11.00.

I'm doing a phd in machine learning in the Traffic Planner project where our main target is to make a tool for rapid congestion prediction based on information about the underlying Oslo map structure and its population with their predicted travel goals. Since map data is well known to be non-Euclidean, I've undertaken the task of learning how we can apply the rather recently developed family of Graph Neural Networks to this problem. Scaling means we can't apply it full scale naively, and so I have made two simpler models I'm using for preliminary application, along with some results using Graph Convolutional Networks.

In this webinar, I will present the Traffic Planner project briefly, then talk about geometric deep learning and GNNs for a bit, and then talk about my own work.

The webinar will be recorded and published after. There will be time for questions after the presentation.

Attend by using this link!


Gaustadalleen 21, 0349 Oslo

Postboks 8600 Majorstua, 0359 Oslo

E-post: toi@toi.no





Nettredaktør: Kommunikasjonsleder Hanne Sparre-Enger