IML Machine Learning Working Group

40/S2-C01 - Salle Curie (CERN)

40/S2-C01 - Salle Curie


Show room on map

Meeting will be by video only on Zoom.

Join Zoom Meeting

Meeting ID: 947 1899 1390
Password:  check email


    • 15:00 15:05
      News 5m
      Speakers: Andrea Wulzer (CERN and EPFL), David Rousseau (LAL-Orsay, FR), Gian Michele Innocenti (CERN), Lorenzo Moneta (CERN), Loukas Gouskos (CERN), Dr Pietro Vischia (Universite Catholique de Louvain (UCL) (BE)), Riccardo Torre (CERN)
    • 15:05 15:35
      VegasFlow and PDFFlow, for Monte Carlo integration/simulation using hardware accelerators including multi-GPU setups 30m

      We present the VegasFlow and PDFFlow packages for fast evaluation of high dimensional integrals based on Monte Carlo integration using dataflow graphs. This software is inspired on the Vegas integration algorithm, ubiquitous in the particle physics community as the driver of cross section integration, and based on Google's powerful TensorFlow library. We benchmark the performance of this library on many different consumer and professional grade GPUs and CPUs, finding up to a 10x improvement with respect to other implementations of the Monte Carlo algorithms considered. Ref:

      Speaker: Stefano Carrazza (CERN)
    • 15:35 16:15
      Exhaustive neural importance sampling applied to Monte Carlo event generation 40m

      The generation of accurate neutrino-nucleus cross-section models needed for neutrino oscillation experiments require simultaneously the description of many degrees of freedom and precise calculations to model nuclear responses. The detailed calculation of complete models makes the Monte Carlo generators slow and impractical. This is a common issue in High Energy Physics event generation. We present Exhaustive Neural Importance Sampling (ENIS), a method based on normalizing flows to find a suitable proposal density for rejection sampling automatically and efficiently, and discuss how this technique solves common issues of the rejection algorithm. Ref: Phys. Rev. D 102, 013003

      Speaker: Sebastian Pina Otey (Universitat Autonoma de Barcelona (ES))
    • 16:15 16:45
      Efficient Event Generation with Normalizing Flows 30m

      With the upcoming HL-LHC, the budget for computing will be insufficient to generate a sufficient amount of Monte-Carlo events for both signal and background predictions. The driving force behind these costs is the inefficiency of the Monte-Carlo phase space generators and the unweighting efficiencies.
      I present i-flow, a Machine Learning code that uses Normalizing Flows for efficient numerical integration and random sampling. I show its performance in comparison to "traditional" algorithms like VEGAS or FOAM for several test cases, including W+jets production with the matrix element generator Sherpa. Ref:,

      Speaker: Claudius Krause (Fermilab)