The Institute of Advanced Research in Artificial Intelligence (IARAI) has launched its 4th annual Traffic4cast competition in collaboration with Here Technologies, the location data and technology platform.
Traffic4cast participants are challenged to utilize the latest in AI methods to model and predict future traffic congestion levels and vehicle speeds across London, Madrid and Melbourne.
The IARAI Traffic4cast competition merges AI with real-world datasets and traffic research to advance the understanding of complex traffic dynamics and systems. Winners will receive their prizes at NeurIPS 2022, the AI conference, and a volume within the Proceedings of Machine Learning will highlight contributions of the Traffic4cast 2022 Competition.
The competition includes two years of real-world data, derived from billions of GPS points from vehicle fleets, provided by Here Technologies. New this year, participants will also leverage data from traffic loop counters embedded in the roadways of London, Madrid, and Melbourne. The aim is to reduce the barriers for using readily available, public loop counter data to predict future traffic state of entire cities.
Here provides participants with traffic movie clips based on two years of real-world data for London, Madrid and Melbourne. The clips were created using Here data based on more than 100 billion GPS probe points from a large fleet of vehicles. The data has been fully anonymized and transformed into high-definition movie clips that, frame by frame, depict traffic over time, including morning, evening, and rush hour traffic events.
Traffic4cast 2022
- Core challenge – participants are asked to predict the congestion level classes (red/yellow/green) for the entire road graph 15 minutes into the future from the past hour of traffic loop counter data only.
- Extended challenge – participants are invited to predict the average speeds on each road segment of the graph 15 minutes into the future from the past hour of traffic loop counter data only.
The winners of the core challenge (congestion classes for the entire road graph) will receive the following prizes:
- 1st place – a voucher or cash prize worth €5,000 to the participant/team plus one free NeurIPS 2022 conference registration.
- 2nd place – a voucher or cash prize worth €3,000 to the participant/team plus one free NeurIPS 2022 conference registration.
- 3rd place – a voucher or cash prize worth €2,000 to the participant/team plus one free NeurIPS 2022 conference registration.
The winners of the extended challenge (average speeds on each road segment) will receive the following prizes:
- 1st place – a voucher or cash prize worth €5,000 to the participant/team plus one free NeurIPS 2022 conference registration.
- 2nd place – a voucher or cash prize worth €3,000 to the participant/team plus one free NeurIPS 2022 conference registration.
- 3rd place – a voucher or cash prize worth €2,000 to the participant/team plus one free NeurIPS 2022 conference registration.
The competition brings together researchers across the world from prominent areas of machine learning, including graph-based modeling, transfer learning, deep learning, and time series prediction.
Sepp Hochreiter, a founding co-director of IARAI and an artificial intelligence pioneer who invented the long short-term memory (LSTM) neural network architecture, said: “Building on the three years of success at NeurIPS 2019–2021, Traffic4cast continues to improve our understanding of complex traffic systems. This year, researchers of modern machine learning will predict traffic congestion in entire cities just from the vehicle counters available on selected points. Besides traffic congestion, advanced models are supposed to also predict the average speeds on a network of roads. Our competition will help to advance and to exploit the latest techniques in machine learning like graph neural networks or physics inspired neural networks.”
“Traffic congestion is a universal challenge that requires deep analysis to understand ‘the hidden rules’ shaping vehicle movements. I am therefore excited to see what predictive models this year’s participants can generate using this expansive new dataset from Here and the latest advances in AI and machine learning,” added Reinhard Köhn, head of research at Here Technologies.
Submissions for this year’s competition are due by October 15, 2022.
