1. General Introduction#
The objective of the course is to be able to use a machine learning approach applied to the treatment of mechanical problems.
Method: computing AI
The adopted perspective is to present the machine learning approach from a numerical methods point of view (using applied numerical methods) rather than from a statistical point of view, which is mainly used by data scientists.
To achieve this, we will use the following approach of “scientific computing,” using the Python programming language commonly used in science.
1.1. Scientific Approach#
Scientific Approach
Physical analysis of the problem
Choice of a mathematical model
Choice of a numerical method
Choice of an algorithmic solution
Programming on a computer
Validation of the approach
Simulation
Analysis of the results
Important
In this approach validation is an essential point of the process !!
Note
The domain of application is (mechanical) engineering modelization, simulation and data analysis. The applicationi of AI to general-purpose language generation (text, image, video) using large language model (LLMs) like ChatGPT is not covered in this course.
1.2. Remarks#
Human expertise
Computers do not have feelings, emotions and cannot duplicate human reasonning. We will always need human experts to analyse and validate the machine learning approach in Science.
Without humans, there is no pilot in the plane !!
1.3. Reference#
Colin Bernet. L’intelligence artificielle: introduction et applications en physique. 2021. URL: https://culturesciencesphysique.ens-lyon.fr/ressource/IA-Bernet-3.xml.
Stephen Boyd and Lieven Vandenberghe. Convex Optimization. Cambridge University Press, 2004.
Steven L. Brunton and J. Nathan Kutz. Chapter 6: Neural Networks and Deep Learning. Cambridge University Press, 2019.
Marc Buffat. Inpros: introduction à la programmation scientifique. 2018. URL: https://perso.univ-lyon1.fr/marc.buffat/2022/BOOK_INPROS/index.html.
Dominique Cardon, Jean-Philippe Cointet, and Antoine Mazières. La revanche des neurones: l’invention des machines inductives et la controverse de l’intelligence artificielle. Réseaux, n° 211(5):173–220, November 2018. doi:10.3917/res.211.0173.
Sébastien Charnoz and Adrian Daerr. Algorithmes de minimisation. 2010. URL: https://irfu.cea.fr/Projets/COAST/methodes_numeriques_MINI.pdf.
Brian M. de Silva, David M. Higdon, Steven L. Brunton, and J. Nathan Kutz. Discovery of physics from data: universal laws and discrepancies. Frontiers in Artificial Intelligence, April 2020. doi:10.3389/frai.2020.00025.
Allen B. Downey. How to think like a computer scientist. 2016. URL: https://www.greenteapress.com/wp/think-python-2e/.
Megan R. Ebers, Katherine M. Steele, and J. Nathan Kutz. Discrepancy modeling framework: learning missing physics, modeling systematic residuals, and disambiguating: between deterministic and random effects. In arXiv. 2023.
Andries P. Engelbrecht. Computational Intelligence: An Introduction (Second Edition). John Wiley and Sons, 2007.
FranceCulture. Apprentissage autosupervisé : ia, au tableau ! 2022. URL: https://www.radiofrance.fr/franceculture/podcasts/la-methode-scientifique/apprentissage-autosupervise-ia-au-tableau-8790358.
Thomas Groensfelder, Fabian Giebeler, Marco Geupel, David Schneider, and Rebecca Jaeger. Application of machine learning procedures for mechanical system modeling: capabilities and caveats to prediction-accuracy. Advanced Modeling and Simulation in Engineering Sciences, June 2020. doi:10.1186/s40323-020-00163-4.
Thomas Groensfelder, Fabian Giebeler, Marco Geupel, David Schneider, and Rebecca Jaeger. Application of machine learning procedures for mechanical system modelling: capabilities and caveats to prediction-accuracy. In Advanced Modeling and Simulation in Engineering Sciences. 2020.
Patrick Hairy. Physics-informed neural networks. 2022. URL: https://metalblog.ctif.com/2022/01/17/physics-informed-neural-networks/.
Mykel J. Kochenderfer and Tim A. Wheeler. Algorithms for Optimization. MIT press, 2019.
J. Ling, K. Zhang, Lu, L., and G. E. Karniadakis. Deep learning of dynamics from data: a neural network approach to understanding chaotic systems. journal of computational physics. Journal of Computational Physics, 2021.
Lu Lu, Xuhui Meng, Zhiping Mao, and and George Em Karniadakis. Deepxde: a deep learning library for solving differential equations. In arXiv. 2020.
MétéoFrance. Initiation au machine learning. 2018. URL: meteofrance/formation-machine-learning.
Andrei Popescu, Seda Polat-Erdeniz, Alexander Felfernig, Mathias Uta, Müslüm Atas, Viet-Man Le, Klaus Pilsl, Martin Enzelsberger, and Thi Ngoc Trang Tran. An overview of machine learning techniques in constraint solving. Journal of Intelligent Information Systems, 58(1):91–118, August 2021. doi:10.1007/s10844-021-00666-5.
M. Raissi, Perdikaris, P., and G. E. Karniadakis. Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational Physics, 2019.
Samuel H. Rudy, J. Nathan Kutz, and Steven L. Brunton. Deep learning of dynamics and signal-noise decomposition with time-stepping constraints. In ArXiv. 2018.
Philippe Spalart. An old-fashioned framework for machine learning in turbulence and modeling. In arXiv. 2022.
Sophie Steger, Franz M. Rohrhofer, and Bernhard C. Geiger. How pinns cheat: predicting chaotic motion of a double pendulum. In 36th Conference on Neural Information Processing Systems. 2022.
Shiliang Sun, Zehui Cao, Han Zhu, and Jing Zhao. A survey of optimization methods a machine and learning perspective. In arXiv. 2019.
Franciszek Szewczyk, Michal Tesnar, Wojciech Trejter, and Wojciech Anyszka. Discovering dynamics, conservation laws and symmetries underlying the double pendulum system: a neural and networks approach. In ArXiv. 2020. doi:ng.
wikipedia. Optimisation (mathématiques). 2024. URL: https://fr.wikipedia.org/wiki/Optimisation_(mathématiques).