* __Informed Machine Learning -- A Taxonomy and Survey of Integrating Knowledge into Learning Systems__
Laura von Rueden, Sebastian Mayer, Katharina Beckh, Bogdan Georgiev, Sven Giesselbach, Raoul Heese, Birgit Kirsch, Julius Pfrommer, Annika Pick, Rajkumar Ramamurthy, Michal Walczak, Jochen Garcke, Christian Bauckhage, Jannis Schuecker
Learning, 18, 2019
https://arxiv.org/pdf/1903.12394.pdf
* __Theory-guided Data Science: A New Paradigm for Scientific Discovery from Data__
Anuj Karpatne, Gowtham Atluri, James H. Faghmous, Michael Steinbach, Arindam Banerjee,
Auroop Ganguly, Shashi Shekhar, Nagiza Samatova, and Vipin Kumar
IEEE Transactions on Knowledge and Data Engineering, 29(10), 2017, 2318-2331
https://arxiv.org/pdf/1612.08544.pdf
* Intro by Karim Mache, JSC, Earth System Data Exploration group
### Discussion
* Olav Zimmermann: There is a good overview in the papers but not a guide which methode to use when. Who has any experience which any of those approaches?
* Stefan Kesselheim: Techniques from Machine Learning Potentials create models that obey basic physical symmetries like permutation invariance and tranlation/rotation invariance of by construction. This suffers from the drawback that features must be hand-crafted again.
* comment Jan Ebert: example for physics informed model. Reinforcement learning example application using an ODE:
https://fluxml.ai/blog/2019/03/05/dp-vs-rl.html
* Esteban Vaca: use advanced simulations as input to the DL model. Question is rather how to make such approaches working
* Discussion on input output relation when incoorporaring prior knowledge, e.g. getting output in the same form/lanuage as the prior
* Ahmed Nebli mentions teacher- student models
https://arxiv.org/abs/1912.13179
* Mehdi points to Cyc: https://en.wikipedia.org/wiki/Cyc
* Olav mentions the discussion on NLP on knowledge based model design versus data driven learning between Gary Marcus an Yann Le Cun, Does AI Need More Innate Machinery: https://www.youtube.com/watch?v=vdWPQ6iAkT4
* Johannes Kruse asks for any eyperience with low fidelity simulations to compensate for too few training data on the complex task
* Stefan asks for groups at FZJ currently working on physics informed
* Edaordo DiNapoli, Esteban Vaca INM-1, Johannes Kruse IEK-STE
### Next meeting
15 February 2021 Explainable Machine Learning
* __Unmasking Clever Hans predictors and assessing what machines really learn__
Sebastian Lapuschkin, Stephan Wäldchen, Alexander Binder, Grégoire Montavon, Wojciech Samek, Klaus-Robert Müller
Nat Commun 10, 1096 (2019)
https://doi.org/10.1038/s41467-019-08987-4
* __Discovering physical concepts with neural networks__
Raban Iten, Tony Metger, Henrik Wilming, Lidia del Rio, Renato Renner
Physical Review Letters, 124(1), 2020, 010508
https://arxiv.org/abs/1807.10300
* Tobias Tesch (IBG-3) agreed to prepare the short intro
* discussion on form of upcoming meetings and organization responsiblities
* shall we use the meetings to let groups introduce their current ML projects?
* Consent: the one introducing the papers, shall shortly introduce his/her own or his/her group's work wrt ML and relation to the papers we discuss
