This is the official announcement of the first « Accuracy\" phase of the TrackML competition (see also https://sites.google.com/site/trackmlparticle/results )
\n\nThe final leaderboard has been confirmed.
\n\nIn 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 (see bios) , has examined the contributions submitted:
\n\nThe 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.
\n\nA big thanks to our sponsors.
\n\nA dedicated 75' TrackML session will take 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. See you there!
\n\nThe second phase of the challenge with a scoring including now a strong incentive on speed is on-going on Codalab till 12th March 2019.
\n\nNewcomers are welcome, especially given that documented software from the top #1 #2 #3 #7 #9 #11 #12 has been made available.
", "messages": [ { "id": "412110", "postDate": "10/29/2018 15:18:48", "content": "This is the official announcement of the first « Accuracy\" phase of the TrackML competition (see also https://sites.google.com/site/trackmlparticle/results )
\n\nThe final leaderboard has been confirmed.
\n\nIn 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 (see bios) , has examined the contributions submitted:
\n\nThe 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.
\n\nA big thanks to our sponsors.
\n\nA dedicated 75' TrackML session will take 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. See you there!
\n\nThe second phase of the challenge with a scoring including now a strong incentive on speed is on-going on Codalab till 12th March 2019.
\n\nNewcomers are welcome, especially given that documented software from the top #1 #2 #3 #7 #9 #11 #12 has been made available.
", "rawMarkdown": "This is the official announcement of the first « Accuracy\" phase of the TrackML competition (see also [https://sites.google.com/site/trackmlparticle/results][1] )\n\nThe [final leaderboard][2] has been confirmed.\n\n - 1st rank : Team top-quarks (Johan Sokrates Wind « icecuber » and Erling Solberg \"erlinsol\" ) wins 12000$ \n - 2nd rank : Team outrunner (Pei Lien Chou) wins 8000$\n - 3rd rank : Team Sergey Gorbunov wins 5000$\n\nIn 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 ([see bios][3]) , has examined the contributions submitted:\n\n - Team «Yuval & Trian» (Yuval Reina «yval_r» and Trian Xylouris «trian2018») , 7th on the final leaderboard, wins the Nvidia V100 for their innovative clustering.\n - 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\n - Team « Finnies» (Nicole Finnie and Liam Finnie ), 12th on the final leaderboard, is invited to NIPS workshop, for their use of a LSTM \n\nThe 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.\n\nA big thanks to [our sponsors][4].\n\nA [dedicated 75' TrackML session][5] will take 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. See you there!\n\nThe second phase of the challenge with a scoring including now a strong incentive on speed is on-going on [Codalab][6] till 12th March 2019.\n\nNewcomers are welcome, especially given that documented software from the top #1 #2 #3 #7 #9 #11 #12 has been made available.\n\n\n [1]: https://sites.google.com/site/trackmlparticle/results\n [2]: https://www.kaggle.com/c/trackml-particle-identification/leaderboard\n [3]: https://sites.google.com/site/trackmlparticle/international-advisory-committee\n [4]: https://www.kaggle.com/c/trackml-particle-identification#About%20the%20Sponsors\n [5]: https://nips.cc/Conferences/2018/Schedule?showEvent=10945\n [6]: https://competitions.codalab.org/competitions/20112", "votes": null }, { "id": "413084", "postDate": "10/31/2018 08:52:28", "content": "Sorry, where are the documented software publicly from the top contestants?
", "rawMarkdown": "Sorry, where are the documented software publicly from the top contestants?", "votes": null }, { "id": "413124", "postDate": "10/31/2018 10:11:18", "content": "On codalab, in the \"participate\" tab (only active if you register): https://competitions.codalab.org/competitions/20112#participate-get_data \nBut here is the direct link: https://drive.google.com/file/d/1iRufZU9vpwbKlRnRXzffCcNS09tkpKr4/view?usp=sharing
", "rawMarkdown": "On codalab, in the \"participate\" tab (only active if you register): https://competitions.codalab.org/competitions/20112#participate-get_data \nBut here is the direct link: https://drive.google.com/file/d/1iRufZU9vpwbKlRnRXzffCcNS09tkpKr4/view?usp=sharing", "votes": null }, { "id": "433908", "postDate": "12/05/2018 16:31:44", "content": "Please check this out:\nhttps://twitter.com/trackmllhc/status/1070339064094736390\nthis is a gif showing that the special score used for the competition is indeed a good proxy : our domain specific plots (particle finding efficiency as a function of various parameters in the ground truth) improve while the score improves\nwe’ll show this at the competition workshop at #NeurIPS this Friday 7th Dec 2018\nhttps://neurips.cc/Conferences/2018/Schedule?showEvent=10945
\n\n(the odd one out: in the bottom left plot, one can briefly see a dark curve rising high, this is Diogo's submission. It shows that his code finds particularly well particles that do NOT come from the z axis (axis of symmetry) (large r0), while most particles come from there (small r0), and most participants focus on particle with small r0, or even assume they come almost exactly from the z axis. Diogo's score is 0.55480, hardly better than the best public kernel, but his algorithm must be quite interesting)
", "rawMarkdown": "Please check this out:\nhttps://twitter.com/trackmllhc/status/1070339064094736390\nthis is a gif showing that the special score used for the competition is indeed a good proxy : our domain specific plots (particle finding efficiency as a function of various parameters in the ground truth) improve while the score improves\nwe’ll show this at the competition workshop at #NeurIPS this Friday 7th Dec 2018\nhttps://neurips.cc/Conferences/2018/Schedule?showEvent=10945\n\n(the odd one out: in the bottom left plot, one can briefly see a dark curve rising high, this is Diogo's submission. It shows that his code finds particularly well particles that do NOT come from the z axis (axis of symmetry) (large r0), while most particles come from there (small r0), and most participants focus on particle with small r0, or even assume they come almost exactly from the z axis. Diogo's score is 0.55480, hardly better than the best public kernel, but his algorithm must be quite interesting)", "votes": null }, { "id": "435161", "postDate": "12/07/2018 15:48:23", "content": "The source code and description of my solution are available here:\nhttps://github.com/diogoff/trackml-100
", "rawMarkdown": "The source code and description of my solution are available here:\nhttps://github.com/diogoff/trackml-100", "votes": null }, { "id": "435180", "postDate": "12/07/2018 16:26:03", "content": "Thanks a lot, this is quite an original approach, which may have applications in slightly different contexts
", "rawMarkdown": "Thanks a lot, this is quite an original approach, which may have applications in slightly different contexts", "votes": null }, { "id": "1149969", "postDate": "01/12/2021 09:16:06", "content": "The final write-up of this Kaggle Accuracy phase for the TrackML challenge is published :
\nS.Amrouche,L.Basara,P.Calafiura,V.Estrade,S.Farrell,D.R. Ferreira, L. Finnie, N. Finnie, C. Germain, V. V. Gligorov, T. Golling, S. Gorbunov, H. Gray, I. Guyon, M. Hushchyn, V. Innocente, M. Kiehn, E. Moyse, J.-F. Puget, Y. Reina, D. Rousseau, A. Salzburger, A. Ustyuzhanin, J.-R. Vlimant, J. S. Wind, T. Xylouris, and Y. Yilmaz, \"The tracking machine learning challenge: Accuracy phase\", in The NeurIPS 2018 Competition, pp. 231–264. Springer International Publishing, Nov., 2019. arXiv:1904.06778 [hep-ex].
\ndoi:10.1007/978-3-030-29135-8_9
The final write-up of the Codalab Throughput phase is being finalized.
", "rawMarkdown": "The final write-up of this Kaggle Accuracy phase for the TrackML challenge is published : \nS.Amrouche,L.Basara,P.Calafiura,V.Estrade,S.Farrell,D.R. Ferreira, L. Finnie, N. Finnie, C. Germain, V. V. Gligorov, T. Golling, S. Gorbunov, H. Gray, I. Guyon, M. Hushchyn, V. Innocente, M. Kiehn, E. Moyse, J.-F. Puget, Y. Reina, D. Rousseau, A. Salzburger, A. Ustyuzhanin, J.-R. Vlimant, J. S. Wind, T. Xylouris, and Y. Yilmaz, \"The tracking machine learning challenge: Accuracy phase\", in The NeurIPS 2018 Competition, pp. 231–264. Springer International Publishing, Nov., 2019. arXiv:[1904.06778](https://arxiv.org/abs/1904.06778) [hep-ex].\n[doi:10.1007/978-3-030-29135-8_9](https://doi.org/10.1007/978-3-030-29135-8_9) \n\nThe final write-up of the Codalab Throughput phase is being finalized.", "votes": null } ], "comments": [ { "id": 1149969, "author_name": "droussea", "author_url": "", "post_date": "01/12/2021 09:16:06", "content": "The final write-up of this Kaggle Accuracy phase for the TrackML challenge is published :
\nS.Amrouche,L.Basara,P.Calafiura,V.Estrade,S.Farrell,D.R. Ferreira, L. Finnie, N. Finnie, C. Germain, V. V. Gligorov, T. Golling, S. Gorbunov, H. Gray, I. Guyon, M. Hushchyn, V. Innocente, M. Kiehn, E. Moyse, J.-F. Puget, Y. Reina, D. Rousseau, A. Salzburger, A. Ustyuzhanin, J.-R. Vlimant, J. S. Wind, T. Xylouris, and Y. Yilmaz, \"The tracking machine learning challenge: Accuracy phase\", in The NeurIPS 2018 Competition, pp. 231–264. Springer International Publishing, Nov., 2019. arXiv:1904.06778 [hep-ex].
\ndoi:10.1007/978-3-030-29135-8_9
The final write-up of the Codalab Throughput phase is being finalized.
", "votes": null, "replies": [] }, { "id": 413084, "author_name": "alankcheung", "author_url": "", "post_date": "10/31/2018 08:52:28", "content": "Sorry, where are the documented software publicly from the top contestants?
", "votes": null, "replies": [] }, { "id": 413124, "author_name": "droussea", "author_url": "", "post_date": "10/31/2018 10:11:18", "content": "On codalab, in the \"participate\" tab (only active if you register): https://competitions.codalab.org/competitions/20112#participate-get_data \nBut here is the direct link: https://drive.google.com/file/d/1iRufZU9vpwbKlRnRXzffCcNS09tkpKr4/view?usp=sharing
", "votes": null, "replies": [] }, { "id": 433908, "author_name": "droussea", "author_url": "", "post_date": "12/05/2018 16:31:44", "content": "Please check this out:\nhttps://twitter.com/trackmllhc/status/1070339064094736390\nthis is a gif showing that the special score used for the competition is indeed a good proxy : our domain specific plots (particle finding efficiency as a function of various parameters in the ground truth) improve while the score improves\nwe’ll show this at the competition workshop at #NeurIPS this Friday 7th Dec 2018\nhttps://neurips.cc/Conferences/2018/Schedule?showEvent=10945
\n\n(the odd one out: in the bottom left plot, one can briefly see a dark curve rising high, this is Diogo's submission. It shows that his code finds particularly well particles that do NOT come from the z axis (axis of symmetry) (large r0), while most particles come from there (small r0), and most participants focus on particle with small r0, or even assume they come almost exactly from the z axis. Diogo's score is 0.55480, hardly better than the best public kernel, but his algorithm must be quite interesting)
", "votes": null, "replies": [ { "id": 435161, "author_name": "diogoff", "author_url": "", "post_date": "12/07/2018 15:48:23", "content": "The source code and description of my solution are available here:\nhttps://github.com/diogoff/trackml-100
", "votes": null, "replies": [] }, { "id": 435180, "author_name": "droussea", "author_url": "", "post_date": "12/07/2018 16:26:03", "content": "Thanks a lot, this is quite an original approach, which may have applications in slightly different contexts
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "412110": "This is the official announcement of the first « Accuracy\" phase of the TrackML competition (see also [https://sites.google.com/site/trackmlparticle/results][1] )\n\nThe [final leaderboard][2] has been confirmed.\n\n - 1st rank : Team top-quarks (Johan Sokrates Wind « icecuber » and Erling Solberg \"erlinsol\" ) wins 12000$ \n - 2nd rank : Team outrunner (Pei Lien Chou) wins 8000$\n - 3rd rank : Team Sergey Gorbunov wins 5000$\n\nIn 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 ([see bios][3]) , has examined the contributions submitted:\n\n - Team «Yuval & Trian» (Yuval Reina «yval_r» and Trian Xylouris «trian2018») , 7th on the final leaderboard, wins the Nvidia V100 for their innovative clustering.\n - 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\n - Team « Finnies» (Nicole Finnie and Liam Finnie ), 12th on the final leaderboard, is invited to NIPS workshop, for their use of a LSTM \n\nThe 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.\n\nA big thanks to [our sponsors][4].\n\nA [dedicated 75' TrackML session][5] will take 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. See you there!\n\nThe second phase of the challenge with a scoring including now a strong incentive on speed is on-going on [Codalab][6] till 12th March 2019.\n\nNewcomers are welcome, especially given that documented software from the top #1 #2 #3 #7 #9 #11 #12 has been made available.\n\n\n [1]: https://sites.google.com/site/trackmlparticle/results\n [2]: https://www.kaggle.com/c/trackml-particle-identification/leaderboard\n [3]: https://sites.google.com/site/trackmlparticle/international-advisory-committee\n [4]: https://www.kaggle.com/c/trackml-particle-identification#About%20the%20Sponsors\n [5]: https://nips.cc/Conferences/2018/Schedule?showEvent=10945\n [6]: https://competitions.codalab.org/competitions/20112", "413084": "Sorry, where are the documented software publicly from the top contestants?", "413124": "On codalab, in the \"participate\" tab (only active if you register): https://competitions.codalab.org/competitions/20112#participate-get_data \nBut here is the direct link: https://drive.google.com/file/d/1iRufZU9vpwbKlRnRXzffCcNS09tkpKr4/view?usp=sharing", "433908": "Please check this out:\nhttps://twitter.com/trackmllhc/status/1070339064094736390\nthis is a gif showing that the special score used for the competition is indeed a good proxy : our domain specific plots (particle finding efficiency as a function of various parameters in the ground truth) improve while the score improves\nwe’ll show this at the competition workshop at #NeurIPS this Friday 7th Dec 2018\nhttps://neurips.cc/Conferences/2018/Schedule?showEvent=10945\n\n(the odd one out: in the bottom left plot, one can briefly see a dark curve rising high, this is Diogo's submission. It shows that his code finds particularly well particles that do NOT come from the z axis (axis of symmetry) (large r0), while most particles come from there (small r0), and most participants focus on particle with small r0, or even assume they come almost exactly from the z axis. Diogo's score is 0.55480, hardly better than the best public kernel, but his algorithm must be quite interesting)", "435161": "The source code and description of my solution are available here:\nhttps://github.com/diogoff/trackml-100", "435180": "Thanks a lot, this is quite an original approach, which may have applications in slightly different contexts", "1149969": "The final write-up of this Kaggle Accuracy phase for the TrackML challenge is published : \nS.Amrouche,L.Basara,P.Calafiura,V.Estrade,S.Farrell,D.R. Ferreira, L. Finnie, N. Finnie, C. Germain, V. V. Gligorov, T. Golling, S. Gorbunov, H. Gray, I. Guyon, M. Hushchyn, V. Innocente, M. Kiehn, E. Moyse, J.-F. Puget, Y. Reina, D. Rousseau, A. Salzburger, A. Ustyuzhanin, J.-R. Vlimant, J. S. Wind, T. Xylouris, and Y. Yilmaz, \"The tracking machine learning challenge: Accuracy phase\", in The NeurIPS 2018 Competition, pp. 231–264. Springer International Publishing, Nov., 2019. arXiv:[1904.06778](https://arxiv.org/abs/1904.06778) [hep-ex].\n[doi:10.1007/978-3-030-29135-8_9](https://doi.org/10.1007/978-3-030-29135-8_9) \n\nThe final write-up of the Codalab Throughput phase is being finalized." }, "source": "meta" }