- author = "Silviu-Marian Udrescu and Max Tegmark",
- title = "{AI Feynman}: A physics-inspired method for symbolic regression",
- journal = "Science Advances",
- year = "2020",
- volume = "6",
- number = "16",
- month = "15 " # apr,
- keywords = "genetic algorithms, genetic programming, research article, computer science",
- ISSN = "2375-2548",
- publisher = "American Association for the Advancement of Science",
-
URL = "
https://advances.sciencemag.org/content/advances/6/16/eaay2631.full.pdf",
-
URL = "
https://pubmed.ncbi.nlm.nih.gov/32426452/",
-
DOI = "
doi:10.1126/sciadv.aay2631",
-
code_url = "https://github.com/SJ001/AI-Feynman",
- size = "17 pages",
- abstract = "A core challenge for both physics and artificial intelligence (AI) is symbolic regression: finding a symbolic expression that matches data from an unknown function. Although this problem is likely to be NP-hard in principle, functions of practical interest often exhibit symmetries, separability, compositionality, and other simplifying properties. In this spirit, we develop a recursive multidimensional symbolic regression algorithm that combines neural network fitting with a suite of physics-inspired techniques. We apply it to 100 equations from the Feynman Lectures on Physics, and it discovers all of them, while previous publicly available software cracks only 71; for a more difficult physics-based test set, we improve the state-of-the-art success rate from 15 to 90percent.",
-
notes = "Is this GP? First tries enumeration of GP like
functions rather than search before other approaches,
eg 6 layer ANN. 'our neural network enables eliminating
variables by discovering symmetries and
separability.'
https://space.mit.edu/home/tegmark/aifeynman.html
Python SymPy package. (Test set) 20 equations, extracted from other seminal physics books. Comparison with Eurequa \cite{Dubcakova:2011:GPEM} \cite{Science09:Schmidt}.
'... in reverse Polish notation (RPN), which would require about 100 times the age of our universe for the brute-force method'.
Replacing (1/4 pi eta) by Coulomb constant. Introducing new operators into brute force search representation based on RPN to speed search.
ANN that can 'repeatedly reduce a problem to simpler ones, eliminating dependent variables by discovering properties such as symmetries and separability in the unknown function'. Additional noise.
See also \cite{LaCava:2021:NeurIPS}
Department of Physics and Center for Brains, Minds and Machines, Massachusetts Institute of Technology",
