Higgs Machine Learning Challenge visits CERN

Europe/Zurich
500-1-001 - Main Auditorium (CERN)

500-1-001 - Main Auditorium

CERN

400
Show room on map
Description

The Higgs Machine Learning challenge (HiggsML https://higgsml.lal.in2p3.fr) ran on the Kaggle platform in summer 2014. Official simulated ATLAS events were publicly released, and participants competed to invent the most powerful algorithms to improve the statistical significance of the Higgs to tau+tau- signal.

The participation was overwhelming with more than 1700 teams, Machine learning specialists, physicists and students, submitted more than 30.000 solutions, making it the most popular Kaggle challenge at the time. During these few months, a typical HEP problem has been tackled with the most advanced machine learning techniques, which have quickly outperformed traditional HEP tools. This mini workshop is a step towards importing the lessons of the challenge back into High Energy Physics.

This event is open to anyone with some interest in advanced machine learning/multivariate analysis techniques.

Videos will be made available publicly on this page within a few days.

A mailing list HEP-data-science@googlegroups.com has just been created to deal with anything concerning both HEP and Data Science, in particular Machine Learning. Announcement/discussion about challenges, workshops, relevant papers, tools, etc... Subscription by sending a mail to HEP-data-science+subscribe@googlegroups.com.

Webcast
There is a live webcast for this event
    • 15:00 15:30
      The Higgs Machine learning challenge 30m
      Overview of the challenge by the organizers. How it was organized, what difficulty had to be overcome, what have we learned so far, next steps.
      Speaker: David Rousseau (LAL-Orsay, FR)
      Slides
      Video in CDS
    • 15:40 16:10
      HEP meets ML award talk : XGBoost 30m
      Tianqi Chen and Tong He (team crowwork) have provided very early in the challenge to all participants XGBoost (for eXtreme Gradient Boosted). It is a parallelised software to train boost decision trees, which has been effectively used by many participants to the challenge. For this, they have won the "HEP meets ML" award which is the invitation to CERN happening today.
      Speakers: Tianqi CHEN (University of Washington, USA) , Tong HE (Simon Fraser University, Canada)
      Slides
      Video in CDS
    • 16:20 16:50
      Challenge winner talk 30m
      Gábor Melis reached the top of the leaderboard using a deep neural network. He will explain all that he did to get there, and, equally interesting, all that he tried which did not work.
      Speaker: Gábor MELIS (Franz Inc., Fixnum Services, Hungary)
      Slides
      Video in CDS
Your browser is out of date!

Update your browser to view this website correctly. Update my browser now

×