Description
The workshop aims at gathering active scientists from different domains (statistics, machine learning, statistical physics, computer science) who made significant contributions in order to address conceptual questions in Data Science, where a unified, well accepted picture is still lacking
 


Invited Speakers:
Afonso S. BANDEIRA,  NYU
Jean BARBIER,  ICTP
Helmut BÖLCSKEI,  ETH Zürich
Simona COCCO,  ENS Paris
Yoshiyuki KABASHIMA,  Tokyo Institute of Technology
Scott KIRKPATRICK,  The Hebrew University of Jerusalem
Yue M. LU,  Harvard
Marc MEZARD,  ENS Paris
Roberto MULET,   Universidad de la Habana
Ilya NEMENMAN,  Emory University
Guillaume OBOZINSKI,  EPFL
Manfred OPPER,  TU Berlin
Olivier PIETQUIN,  Google Brain
Federico RICCI-TERSENGHI,  University of Rome La Sapienza
Andrew SAXE,  University of Oxford
Jared TANNER, University of Oxford
Naftali TISHBY,  Hebrew University of Jerusalem
Lenka ZDEBOROVA,  CNRS
Riccardo ZECCHINA,  Bocconi U., Milan
Hai-Jun ZHOU,  Chinese Academy of Sciences


 

 
Go to day
  • Monday, 30 September 2019
    • 08:00 - 20:00 Monday, 30 September
      • 08:00 Registration and administrative formalities 1h0' ( Leonardo Building - Lobby )
      • 08:45 Welcome address by the Organisers and ICTP Director 15'
      • 09:00 Machine learning with neural networks: the importance of data structure 45' ( Leonardo Building - Budinich Lecture Hall )
        Speaker: Marc MEZARD (ENS, Paris)
        Material: Video
      • 09:45 Statistical-to-computational gaps in high dimensional hypothesis testing and other problems 45'
        Speaker: Afonso S. BANDEIRA (NYU)
        Material: Video
      • 10:30 Coffee break 30' ( Leonardo Building - Terrace )
      • 11:00 Recent advances in the information theory of deep neural networks and the computational benefits of the hidden layers 45'
        Speaker: Naftali TISHBY (Hebrew U., Jerusalem)
        Material: Video
      • 11:45 TBA 45' ( Leonardo Building - Budinich Lecture Hall )
        Speaker: Riccardo ZECCHINA (Bocconi U., Milan)
        Material: Video
      • 12:30 Lunch break 3h0'
      • 15:30 Analysing the dynamics of message passing algorithms for inference using statistical mechanics and random matrix theory 45'
        Speaker: Manfred OPPER (TU Berliin)
        Material: Video
      • 16:15 From sequence data to protein design with Boltzmann machines 45'
        Speaker: Simona COCCO (ENS, Paris)
        Material: Video
      • 19:00 Welcome get-together 1h0' ( Adriatico Guest House, Terrace )
  • Tuesday, 1 October 2019
    • 09:00 - 16:00 Tuesday, 1 October
      • 09:00 The statistical significance of perfect linear separation 45'
        Speaker: Jared TANNER (Oxford U.)
        Material: Video
      • 09:45 Fundamental limits of deep neural network learning 45' ( Leonardo Building - Budinich Lecture Hall )
        Speaker: Helmut BÖLCSKEI (ETH Zürich)
        Material: Video
      • 10:30 Group Photo 5' ( Lobby, Leonardo Building )
      • 10:30 Coffee break 30' ( Leonardo Building - Terrace )
      • 11:00 Training and generalization dynamics in simple deep networks 45' ( Leonardo Building - Budinich Lecture Hall )
        Speaker: Andrew SAXE (Oxford U.)
        Material: Video
      • 11:45 Perspectives on algorithm design for data science 45'
        Speaker: Joachim M. BUHMANN (ETH Zurich)
        Material: Video
      • 12:30 Lunch break 1h30'
      • 14:00 Presentation of Participants' posters and coffee break 2h0' ( Gallery behind Budinich Lecture Hall (LB) )
  • Wednesday, 2 October 2019
    • 09:00 - 12:30 Wednesday, 2 October
      • 09:00 Analysing the gradient descent-based dynamics in complex landscapes 45'
        Speaker: Lenka ZDEBOROVA (CNRS)
        Material: Video
      • 09:45 Semi-analytic resampling in generalized linear models 45' ( Leonardo Building - Budinich Lecture Hall )
        Speaker: Yoshiyuki KABASHIMA (Tokyo Institute of Technology)
        Material: Reading material Video
      • 10:30 Coffee break 30' ( Leonardo Building - Terrace )
      • 11:00 Exploiting beliefs out of equilibrium 45'
        Speaker: Roberto MULET (Universidad de la Habana)
        Material: Video
      • 11:45 Deep reinforcement learning with demonstrations 45'
        Speaker: Olivier PIETQUIN (Google Brain)
  • Thursday, 3 October 2019
    • 09:00 - 13:15 Thursday, 3 October
      • 09:00 Spectral methods for high-dimensional estimation: Asymptotics and fundamental limits 45'
        Speaker: Yue M. LU (Harvard U.)
        Material: Video
      • 09:45 Intrinsic dimension of data representations in deep neural networks 45' ( Leonardo Building - Budinich Lecture Hall )
        Speaker: Davide ZOCCOLAN (SISSA - International School for Advanced Studies)
        Material: Video
      • 10:30 Coffee break 30' ( Leonardo Building - Terrace )
      • 11:00 Learning interpretable models of dynamics of living systems 45'
        Speaker: Ilya NEMENMAN (Emory U.)
        Material: Video
      • 11:45 Statistical mechanics of random optimization & inference problems: from phase transitions to algorithmic behavior 45'
        Speaker: Federico RICCI-TERSENGHI (U. of Rome, La Sapienza)
        Material: Video
  • Friday, 4 October 2019
    • 09:00 - 12:30 Friday, 4 October
      • 09:00 Deep search without early information 45' ( Leonardo Building - Budinich Lecture Hall )
        Speaker: Scott KIRKPATRICK (Hebrew U., Jerusalem)
        Material: Video
      • 09:45 Beyond the thermodynamic limit in high-dimensional inference: finite size corrections and the «all-or-nothing» phenomenon 45'
        Speaker: Jean BARBIER (ICTP)
        Material: Video
      • 10:30 Coffee break 30' ( Leonardo Building - Terrace )
      • 11:00 An SDCA-powered inexact dual augmented Lagrangian method for fast CRF learning 45'
        Speaker: Guillaume OBOZINSKI (EPFL, Lausanne)
        Material: Video
      • 11:45 Compressed sensing and optimal perceptron learning: Some ideas based on heuristic iteration 45'
        Speaker: Hai-Jun ZHOU (Chinese Academy of Sciences)
        Material: Video