Traffic4cast 2022 competitors calls on machine studying

  • Individuals tasked to mannequin and forecast site visitors congestion throughout 3 world cities
  • Represents a singular real-world, graph-based prediction drawback
  • Winners to be offered on the prestigious NeurIPS 2022 convention

August 2, 2022

Vienna – The Institute of Superior Analysis in Synthetic Intelligence (IARAI), an impartial world machine studying analysis institute, has introduced the opening of its 4th yearly Traffic4cast competitors in collaboration with HERE Applied sciences. Individuals are challenged to make the most of the most recent in AI strategies to mannequin and predict future site visitors congestion ranges and automobile speeds throughout London, Madrid and Melbourne.

The IARAI Traffic4cast competitors is exclusive in merging AI with real-world datasets and site visitors analysis to advance the understanding of advanced site visitors dynamics and methods. Winners will obtain prizes at NeurIPS 2022, the main convention in AI, and a quantity throughout the Proceedings of Machine Studying Analysis will spotlight contributions of the Traffic4cast 2022 Competitors.

The competitors consists of two years of real-world information supplied by HERE Applied sciences, derived from billions of GPS factors from automobile fleets. New this yr, contributors will leverage information from site visitors loop counters embedded within the roadways of London, Madrid and Melbourne. The goal is to cut back the boundaries for utilizing available, public loop counter information to foretell the longer term site visitors state of complete cities.

Traffic4cast 2022

  • Core problem – contributors are requested to foretell the cingestion stage lessons (pink/yellow/inexperienced) for your complete highway graph quarter-hour into the longer term from the previous hour of site visitors loop counter information solely.
  • Prolonged problem – contributors are invited to foretell the common speeds on every highway section of the graph quarter-hour into the longer term from the previous hour of site visitors loop counter information solely.

HERE will present contributors with site visitors film clips primarily based on two years of real-world information for London, Madrid and Melbourne. The clips have been created utilizing HERE information primarily based on greater than 100 billion GPS probe factors from a big fleet of automobiles. The information has been totally anonymized and remodeled into high-definition film clips that, body by body, depict site visitors over time, together with morning, night and rush hour site visitors occasions.

The competitors brings collectively researchers throughout the globe from a number of distinguished areas of machine studying, together with graph-based modeling, switch studying, deep studying, and time collection prediction. The competitors final yr targeted on the time-domain shift in site visitors because of COVID-19.

Sepp Hochreiter, a founding co-director of IARAI and a man-made intelligence pioneer, who invented the lengthy short-term reminiscence (LSTM) neural community structure, stated: “Constructing on the three years of success at NeurIPS 2019–2021, Traffic4cast continues to enhance our understanding of advanced site visitors methods. This yr, researchers of recent machine studying will predict site visitors congestion in complete cities simply from the automobile counters accessible at chosen factors. Moreover site visitors congestion, superior fashions are speculated to additionally predict the common pace on a community of roads. Our competitors will assist to advance and to use the most recent methods in machine studying like graph neural networks or physics impressed neural networks.”

“Site visitors congestion is a common problem that requires deep evaluation to know ‘the hidden guidelines’ shaping automobile actions. I’m due to this fact excited to see what predictive fashions this yr’s contributors can generate utilizing this expansive new dataset from HERE and the most recent advances in AI and machine studying,” stated Reinhard Köhn, Head of Analysis at HERE Applied sciences.

The annual Traffic4cast competitors receives a whole bunch of submissions from competing groups all over the world, with previous winners from South Korea, the US, China and Sweden.

The winners of the core problem (congestion lessons for your complete highway graph) will obtain the next prizes:

  • 1st place – a voucher or money prize price 5,000 EUR to the participant/workforce plus one free NeurIPS 2022 convention registration.
  • 2nd place – a voucher or money prize price 3,000 EUR to the participant/workforce plus one free NeurIPS 2022 convention registration.
  • third place – a voucher or money prize price 2,000 EUR to the participant/workforce plus one free NeurIPS 2022 convention registration.

The winners of the prolonged problem (common speeds on every highway section) will obtain the next prizes:

  • 1st place – a voucher or money prize price 5,000 EUR to the participant/workforce plus one free NeurIPS 2022 convention registration.
  • 2nd place – a voucher or money prize price 3,000 EUR to the participant/workforce plus one free NeurIPS 2022 convention registration.
  • third place – a voucher or money prize price 2,000 EUR to the participant/workforce plus one free NeurIPS 2022 convention registration.

Submissions for this yr’s competitors are due by October 15, 2022. To study extra and compete in Traffic4Cast 2022, click on right here.

Media Contact
Dr. Sebastian Kurme
+49 173 515 3549
[email protected] here.com

Jordan Stark
+1 312 316 4537
[email protected] here.com

About HERE Applied sciences
HERE, a location information and know-how platform, strikes individuals, companies and cities ahead by harnessing the ability of location. By leveraging our open platform, we empower our prospects to realize higher outcomes – from serving to a metropolis handle its infrastructure or a enterprise to optimize its property to guiding drivers to their vacation spot safely. To study extra about HERE, please go to right here.com and 360.right here.com.

.