A Comparison of Mesh-Free Differentiable Programming and Data-Driven Strategies for Optimal Control under PDE Constraints (2310.02286v1)
Abstract: The field of Optimal Control under Partial Differential Equations (PDE) constraints is rapidly changing under the influence of Deep Learning and the accompanying automatic differentiation libraries. Novel techniques like Physics-Informed Neural Networks (PINNs) and Differentiable Programming (DP) are to be contrasted with established numerical schemes like Direct-Adjoint Looping (DAL). We present a comprehensive comparison of DAL, PINN, and DP using a general-purpose mesh-free differentiable PDE solver based on Radial Basis Functions. Under Laplace and Navier-Stokes equations, we found DP to be extremely effective as it produces the most accurate gradients; thriving even when DAL fails and PINNs struggle. Additionally, we provide a detailed benchmark highlighting the limited conditions under which any of those methods can be efficiently used. Our work provides a guide to Optimal Control practitioners and connects them further to the Deep Learning community.
- Felipe Miguel Aparicio Acosta. 1995. Radial basis function and related models: an overview. Signal Processing 45, 1 (1995), 37–58.
- Grégoire Allaire. 2015. A review of adjoint methods for sensitivity analysis, uncertainty quantification and optimization in numerical codes. Ingénieurs de l’Automobile 836 (2015), 33–36.
- Coexistence and optimal control problems for a degenerate predator–prey model. Journal of mathematical analysis and applications 378, 2 (2011), 528–540.
- Hazem Ali Attia and NA Kotb. 1996. MHD flow between two parallel plates with heat transfer. Acta mechanica 117, 1-4 (1996), 215–220.
- Automatic differentiation in machine learning: a survey. Journal of Marchine Learning Research 18 (2018), 1–43.
- Deep learning-based schemes for singularly perturbed convection-diffusion problems. ESAIM: Proceedings and Surveys 73 (8 2023), 48–67. https://doi.org/10.1051/proc/202373048
- JAX-Fluids: A fully-differentiable high-order computational fluid dynamics solver for compressible two-phase flows. Computer Physics Communications 282 (2023), 108527.
- Mécanique céleste et contrôle des véhicules spatiaux. Vol. 51. Springer Science & Business Media.
- JAX: composable transformations of Python+NumPy programs. http://github.com/google/jax
- E.W. Cheney. 1966. Introduction to Approximation Theory. McGraw-Hill Book Company. https://books.google.co.uk/books?id=9uZQAAAAMAAJ
- Alexandre Joel Chorin. 1967. The numerical solution of the Navier-Stokes equations for an incompressible fluid. Bull. Amer. Math. Soc. 73, 6 (1967), 928–931.
- Scientific machine learning through physics–informed neural networks: Where we are and what’s next. Journal of Scientific Computing 92, 3 (2022), 88.
- Bengt Fornberg and Cécile Piret. 2008. A stable algorithm for flat radial basis functions on a sphere. SIAM Journal on Scientific Computing 30, 1 (2008), 60–80.
- Bengt Fornberg and Julia Zuev. 2007. The Runge phenomenon and spatially variable shape parameters in RBF interpolation. Computers & Mathematics with Applications 54, 3 (2007), 379–398.
- Yarin Gal and Zoubin Ghahramani. 2016. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. In international conference on machine learning. PMLR, 1050–1059.
- Christophe Geuzaine and Jean-François Remacle. 2009. Gmsh: A 3-D finite element mesh generator with built-in pre- and post-processing facilities. Internat. J. Numer. Methods Engrg. 79, 11 (2009), 1309–1331. https://doi.org/10.1002/nme.2579 arXiv:https://onlinelibrary.wiley.com/doi/pdf/10.1002/nme.2579
- Michael B Giles and Niles A Pierce. 2000. An introduction to the adjoint approach to design. Flow, turbulence and combustion 65 (2000), 393–415.
- Andreas Griewank and Andrea Walther. 2008. Evaluating derivatives: principles and techniques of algorithmic differentiation. SIAM.
- Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems 25, 1106-1114 (2012), 1.
- Learning to Control PDEs with Differentiable Physics. (1 2020). http://arxiv.org/abs/2001.07457
- Multilayer feedforward networks are universal approximators. Neural networks 2, 5 (1989), 359–366.
- Dojo: A differentiable simulator for robotics. arXiv preprint arXiv:2203.00806 9 (2022).
- Edward J Kansa. 1990. Multiquadrics—A scattered data approximation scheme with applications to computational fluid-dynamics—I surface approximations and partial derivative estimates. Computers & Mathematics with applications 19, 8-9 (1990), 127–145.
- Patrick Kidger. 2022. On Neural Differential Equations. arXiv:2202.02435 [cs.LG]
- Harold W Kuhn and Albert W Tucker. 1951. Nonlinear programming. In Proceedings of the 2nd Berkeley Symposium on Mathematics, Statistics and Probability. University of California Press, Berkeley, 481–492.
- Joseph Louis Lagrange. 1853. Mécanique analytique. Vol. 1. Mallet-Bachelier.
- A comparison of automatic differentiation and continuous sensitivity analysis for derivatives of differential equation solutions. In 2021 IEEE High Performance Extreme Computing Conference (HPEC). IEEE, 1–9.
- Saviz Mowlavi and Saleh Nabi. 2023. Optimal control of PDEs using physics-informed neural networks. J. Comput. Phys. 473 (2023), 111731.
- Nonlinear optimal control strategies for buoyancy-driven flows in the built environment. Computers & Fluids 194 (2019), 104313.
- Mitsuba 2: A retargetable forward and inverse renderer. ACM Transactions on Graphics (TOG) 38, 6 (2019), 1–17.
- Roussel Nzoyem. [n. d.]. Updec. https://github.com/ddrous/Updec/ Third Updec pre-release version..
- Mark JL Orr et al. 1996. Introduction to radial basis function networks.
- Jooyoung Park and Irwin W Sandberg. 1991. Universal approximation using radial-basis-function networks. Neural computation 3, 2 (1991), 246–257.
- The maximum principle. The Mathematical Theory of Optimal Processes. New York: John Wiley and Sons (1962).
- Michael James David Powell. 1981. Approximation theory and methods. Cambridge university press.
- Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational physics 378 (2019), 686–707.
- Hidden fluid mechanics: A Navier-Stokes informed deep learning framework for assimilating flow visualization data. arXiv preprint arXiv:1808.04327 (2018).
- Learning representations by back-propagating errors. nature 323, 6088 (1986), 533–536.
- Amit Sabne. 2020. Xla: Compiling machine learning for peak performance. (2020).
- Samuel S Schoenholz and Ekin D Cubuk. 2019. Jax md: End-to-end differentiable, hardware accelerated, molecular dynamics in pure python. (2019).
- A High-Order Accurate Meshless Method for Solution of Incompressible Fluid Flow Problems. (10 2020). https://doi.org/10.1016/j.jcp.2021.110623
- Shantanu Shahane and Surya Pratap Vanka. 2023. A semi-implicit meshless method for incompressible flows in complex geometries. J. Comput. Phys. 472 (2023), 111715.
- Differentiable modelling to unify machine learning and physical models for geosciences. Nature Reviews Earth & Environment (2023), 1–16.
- Andrei I Tolstykh. 2000. On using RBF-based differencing formulas for unstructured and mixed structured-unstructured grid calculations. In Proceedings of the 16th IMACS world congress, Vol. 228. Lausanne, 4606–4624.
- Emmanuel Trélat. 2005. Contrôle optimal: théorie & applications. Vol. 36. Vuibert Paris.
- Modeling, state estimation, and optimal control for the US COVID-19 outbreak. Scientific reports 10, 1 (2020), 10711.
- Guido Van Rossum et al. 2007. Python Programming Language.. In USENIX annual technical conference, Vol. 41. Santa Clara, CA, 1–36.
- Attention is all you need. Advances in neural information processing systems 30 (2017).
- Paul Werbos. 1974. Beyond regression: New tools for prediction and analysis in the behavioral sciences. PhD thesis, Committee on Applied Mathematics, Harvard University, Cambridge, MA (1974).
- Paul J Werbos. 1990. Backpropagation through time: what it does and how to do it. Proc. IEEE 78, 10 (1990), 1550–1560.
- R. Zamolo and E. Nobile. 2019. Solution of incompressible fluid flow problems with heat transfer by means of an efficient RBF-FD meshless approach. Numerical Heat Transfer, Part B: Fundamentals 75 (1 2019), 19–42. Issue 1. https://doi.org/10.1080/10407790.2019.1580048