The competitions has run in two phases

Throughput Phase

This is the official annnouncement of the "Throughput" phase of the TrackML competition. The final leaderboard is:

• 1st rank : sgorbuno (Sergei Gorbunov) wins 7k$
• 2nd rank : fastrack (Dmitry Emelyanov) wins 5k$
• 3rd rank : cloudkitchen (Marcel Kunze) wins 3k$

They were able to maintain high score with speed well below 10s per event (even below 1s for the first two), well below the state of the art.

They have been invited to participate to the TrackML Grand Finale workshop at CERN in July 2019, were they reported on their techniques. A write-up is in preparation.

A big thanks to our sponsors

Accuracy Phase

This is the official announcement of the first « Accuracy" phase of the TrackML competition

The final leaderboard https://www.kaggle.com/c/trackml-particle-identification/leaderboard has been confirmed

• 1st rank : Team top-quarks (Johan Sokrates Wind « icecuber » and Erling Solberg "erlinsol" ) wins 12000$
• 2nd rank : outrunner (Pei Lien Chou) wins 8000$
• 3rd rank : Sergey Gorbunov wins 5000$

In additional, the TrackML International Advisory Committee composed of High Energy Physicists and Computer Scientists Markus Elsing, Frank Gaede, Alison Lowndes, Maurizio Pierini, Danilo Rezende and Marc Schoenauer (as detailed in https://sites.google.com/site/trackmlparticle/international-advisory-committee), has examined the contributions submitted:

• Team «Yuval & Trian» (Yuval Reina «yval_r» and Trian Xylouris «trian2018») , 7th on the final leaderboard, wins the Nvidia V100 for their innovative clustering.
• Team « DBSCAN for ever » (Jean-Francois Puget « CPMP »), 9th on the leaderboard is invited to CERN dedicated workshop in 2019, for pushing DBSCAN to the limit
• Team « Finnies» (Nicole Finnie and Liam Finnie ), 12th on the final leaderboard, is invited to NIPS workshop, for their use of a LSTM

The proposed methods clearly demonstrate a necessity of further synergy between model-based and data-based approaches. The organizing committee and international advisory committee are glad to acknowledge that the winning solutions tried to elaborate along this direction. Being original and innovative the proposed methods attempted at incorporating the best from physical models and machine learning targeting a trade off between the performance and complexity. We do believe that the experience and lessons learned in the TrackML challenge will further stimulate a dialog between these communities and will be mutually beneficial and enriching.

A big thanks to our sponsors.

A dedicated 75' TrackML session has taken place at Montreal NIPS 2018 Competition workshop Friday 7th December , with talks from Andreas Salzburger and David Rousseau for the organisers, Johan Sokrates Wind "icecuber", Nicole and Liam Finnies, and Sergei Gorbunov: https://nips.cc/Conferences/2018/Schedule?showEvent=10945 . A chapter for the NeurIPS 2018 Competition Book has been accepted for publication.