Workshop on Artificial Scientific Discovery 2021

Please note that this website refers to a past event - for reference use only

 

This workshop aims to bring together some of the world's experts working on the most exciting routes towards true automated scientific discovery. The explicit purpose of this workshop is to emphasize those cases where the machine generates new insights on its own (rather than merely exploiting the well-known tools of machine learning as an aid in human-directed discovery). The most important goal of this workshop is to inspire discussions on possible avenues for future developments.

The workshop took place online, from June 29 (Tue) to July 1 (Thu), running each day from 4pm CET (=10 am EDT) to about 8pm, including a break.

 

Organizers
Leopoldo Sarra and Florian Marquardt (Max Planck Institute for the Science of Light, Theory Division)



Scientific Program

Tuesday,  June 29th

               Introduction  Recording
4:00 pm  Ross King (Chalmers): The Automation of Science    Slides    Recording
4:40 pm  Hans Briegel (Innsbruck): Projective simulation as an interpretable AI for basic science    Slides
5:20 pm  (break)
5:40 pm  Natalia Ares (Oxford): From chips to qubits faster than human experts    Slides    Recording
6:20 pm  James P. Sethna (Cornell): Sloppy models, differential geometry, and why science works     Slides    Recording
7:00 pm  J. Nathan Kutz (U of Washington, Seattle): Deep Learning for the Discovery of Coordinates and Dynamics    Slides    Recording

Wednesday,  June 30th

4:00 pm  Renato Renner (ETH Zürich): Discovering physical concepts using neural networks    Slides    Recording
4:40 pm  Mario Krenn (Toronto): A journey from computer-designed Quantum Experiments to computer-inspired Scientific Understanding    Slides    Recording
5:20 pm  (workshop group picture and break)
5:40 pm  Sebastian Wetzel (Perimeter Institute, Canada): Interpreting artificial neural networks in the context of theoretical physics    Recording
6:20 pm  Tailin Wu (Stanford): Machine learning of physics theories and its universal tradeoff between accuracy and simplicity     Slides   Recording
7:00 pm  POSTER SESSION (open ended)

Thursday,  July 1st

4:00 pm  Kevin Ellis (Cornell): The Role of Higher-level Knowledge in Discovery Problems: Programs and Hierarchical Bayes    Slides    Recording
4:40 pm  Georg Martius (MPI for Intelligent Systems, Tübingen): Discovering Equations using Machine Learning    Slides    Recording
5:20 pm  (break)
5:40 pm  Peter Battaglia (Deep Mind): Physical inductive biases for learning simulation and scientific discovery    Recording
6:20 pm  PANEL DISCUSSION and wrap-up (90 mins.)


Invited Speakers

  • Natalia Ares (Oxford): From chips to qubits faster than human experts
  • Peter Battaglia (Deep Mind): Physical inductive biases for learning simulation and scientific discovery
  • Hans Briegel (Innsbruck): Projective simulation as an interpretable AI for basic science
  • Kevin Ellis (Cornell): The Role of Higher-level Knowledge in Discovery Problems: Programs and Hierarchical Bayes
  • Ross King (Chalmers): The Automation of Science
  • Mario Krenn (Toronto): A journey from computer-designed Quantum Experiments to computer-inspired Scientific Understanding
  • J. Nathan Kutz (U of Washington, Seattle): Deep Learning for the Discovery of Coordinates and Dynamics
  • Georg Martius (MPI for Intelligent Systems, Tübingen): Discovering Equations using Machine Learning
  • Renato Renner (ETH Zürich): Discovering physical concepts using neural networks
  • James P. Sethna (Cornell): Sloppy models, differential geometry, and why science works
  • Sebastian Wetzel (Perimeter Institute, Canada): Interpreting artificial neural networks in the context of theoretical physics
  • Tailin Wu (Stanford): Machine learning of physics theories and its universal tradeoff between accuracy and simplicity

Contributed posters

Besides the invited talks, there will also be a virtual poster session, where we invite contributed presentations from anyone. Posters can cover a wide range of topics, as long as they are still related to machine learning or computer-enhanced discovery in the natural sciences and engineering.

Poster submission deadline: June 25 (Friday) - please submit a link to a PDF of your poster (single page) to gesine.murphy@mpl.mpg.de , e.g. by uploading it to Dropbox, ownCloud, or a similar service.


Location of the participants:


MPL Research Centers and Schools