Details
| Original language | English |
|---|---|
| Article number | 010508 |
| Journal | Physical review letters |
| Volume | 124 |
| Issue number | 1 |
| Publication status | Published - 8 Jan 2020 |
| Externally published | Yes |
Abstract
Despite the success of neural networks at solving concrete physics problems, their use as a general-purpose tool for scientific discovery is still in its infancy. Here, we approach this problem by modeling a neural network architecture after the human physical reasoning process, which has similarities to representation learning. This allows us to make progress towards the long-term goal of machine-assisted scientific discovery from experimental data without making prior assumptions about the system. We apply this method to toy examples and show that the network finds the physically relevant parameters, exploits conservation laws to make predictions, and can help to gain conceptual insights, e.g., Copernicus' conclusion that the solar system is heliocentric.
ASJC Scopus subject areas
Cite this
- Standard
- Harvard
- Apa
- Vancouver
- BibTeX
- RIS
In: Physical review letters, Vol. 124, No. 1, 010508, 08.01.2020.
Research output: Contribution to journal › Article › Research › peer review
}
TY - JOUR
T1 - Discovering Physical Concepts with Neural Networks
AU - Iten, Raban
AU - Metger, Tony
AU - Wilming, Henrik
AU - Del Rio, Lídia
AU - Renner, Renato
N1 - Funding Information: We would like to thank Alessandro Achille, Serguei Beloussov, Ulrich Eberle, Thomas Frerix, Viktor Gal, Thomas Häner, Maciej Koch-Janusz, Aurelien Lucchi, Ilya Nemenman, Joseph M. Renes, Andrea Rocchetto, Norman Sieroka, Ernest Y.-Z. Tan, Jinzhao Wang, and Leonard Wossnig for helpful discussions. We acknowledge support from the Swiss National Science Foundation through SNSF Project No. 200020_165843, the Swiss National Supercomputing Centre (CSCS) under project ID da04, and through the National Centre of Competence in Research Quantum Science and Technology (QSIT). L. d.R. and R. R. furthermore acknowledge support from the FQXi grant Physics of the observer. T. M. acknowledges support from ETH Zürich and the ETH Foundation through the Excellence Scholarship & Opportunity Programme.
PY - 2020/1/8
Y1 - 2020/1/8
N2 - Despite the success of neural networks at solving concrete physics problems, their use as a general-purpose tool for scientific discovery is still in its infancy. Here, we approach this problem by modeling a neural network architecture after the human physical reasoning process, which has similarities to representation learning. This allows us to make progress towards the long-term goal of machine-assisted scientific discovery from experimental data without making prior assumptions about the system. We apply this method to toy examples and show that the network finds the physically relevant parameters, exploits conservation laws to make predictions, and can help to gain conceptual insights, e.g., Copernicus' conclusion that the solar system is heliocentric.
AB - Despite the success of neural networks at solving concrete physics problems, their use as a general-purpose tool for scientific discovery is still in its infancy. Here, we approach this problem by modeling a neural network architecture after the human physical reasoning process, which has similarities to representation learning. This allows us to make progress towards the long-term goal of machine-assisted scientific discovery from experimental data without making prior assumptions about the system. We apply this method to toy examples and show that the network finds the physically relevant parameters, exploits conservation laws to make predictions, and can help to gain conceptual insights, e.g., Copernicus' conclusion that the solar system is heliocentric.
UR - http://www.scopus.com/inward/record.url?scp=85078230753&partnerID=8YFLogxK
U2 - 10.1103/PhysRevLett.124.010508
DO - 10.1103/PhysRevLett.124.010508
M3 - Article
C2 - 31976717
AN - SCOPUS:85078230753
VL - 124
JO - Physical review letters
JF - Physical review letters
SN - 0031-9007
IS - 1
M1 - 010508
ER -