Related references
Note: Only part of the references are listed.
Article
Mathematics, Applied
Siddhartha Mishra et al.
Summary: Physics-informed neural networks (PINNs) have gained popularity for accurately approximating solutions of partial differential equations (PDEs). This paper provides upper bounds for the generalization error of PINNs in solving the forward problem of PDEs. By introducing an abstract formalism and utilizing the stability properties of the underlying PDE, an estimate for the generalization error is derived in terms of the training error and number of training samples. The proposed theory is validated through numerical experiments on various nonlinear PDEs.
IMA JOURNAL OF NUMERICAL ANALYSIS
(2023)
Article
Mathematics, Applied
T. De Ryck et al.
Summary: We provide rigorous bounds on the errors of approximating the incompressible Navier-Stokes equations using (extended) physics-informed neural networks. We prove that the residual of the underlying partial differential equation can be made arbitrarily small for tanh neural networks with two hidden layers. Furthermore, the total error can be estimated based on the training error, network size, and number of quadrature points. The theory is illustrated through numerical experiments.
IMA JOURNAL OF NUMERICAL ANALYSIS
(2023)
Article
Computer Science, Interdisciplinary Applications
Sifan Wang et al.
Summary: This work investigates the Neural Tangent Kernel (NTK) of Physics-informed neural networks (PINNs) and demonstrates that it can converge to a deterministic kernel that remains constant during training under appropriate conditions. A novel gradient descent algorithm is proposed to adaptively calibrate the convergence rate of total training error using the eigenvalues of NTK. A series of numerical experiments are conducted to validate the theory and practical effectiveness of the proposed algorithms.
JOURNAL OF COMPUTATIONAL PHYSICS
(2022)
Article
Computer Science, Interdisciplinary Applications
Ravi G. Patel et al.
Summary: Using a least squares space-time control volume scheme can more naturally handle regularity requirements, imposition of boundary conditions, entropy compatibility, and conservation, reducing requisite hyperparameters. This framework introduces inductive biases to discover thermodynamically consistent equations of state guaranteeing hyperbolicity, providing a method for discovering continuum shock models.
JOURNAL OF COMPUTATIONAL PHYSICS
(2022)
Article
Energy & Fuels
Muhammad M. Almajid et al.
Summary: The paper explores the use of Physics Informed Neural Networks (PINN) to model the Buckley-Leverett problem, showing the importance of coupling between observed data and physics-informed neural networks in different parameter spaces. The results demonstrate that PINNs can capture overall trends without observed data, but greatly improve resolution and accuracy with observed data, and perform better than Artificial Neural Networks (ANN) when early- and late-time behavior data are included.
JOURNAL OF PETROLEUM SCIENCE AND ENGINEERING
(2022)
Article
Physics, Multidisciplinary
Yifan Mo et al.
Summary: In this work, a deep learning algorithm for vector solitons of the coupled nonlinear Schrodinger equation (CNLSE) is explored. By improving the training algorithm, the proposed method can better approximate the dynamical behaviors of solitons and achieve faster convergence rate.
Article
Computer Science, Information Systems
Muhammad Rafiq et al.
Summary: Solving parametric partial differential equations using artificial intelligence allows for faster convergence and higher accuracy compared to traditional numerical solvers.
Article
Statistics & Probability
Cynthia Rudin et al.
Summary: This work highlights the fundamental principles of interpretable machine learning and identifies 10 technical challenge areas in this field, including optimizing sparse models, scoring systems, and adding constraints for better interpretability. It serves as a useful starting point for statisticians and computer scientists interested in interpretable machine learning.
STATISTICS SURVEYS
(2022)
Article
Water Resources
Gege Wen et al.
Summary: Numerical simulation of multiphase flow plays a vital role in geoscience applications. The U-FNO neural network architecture, based on the Fourier neural operator (FNO), offers a superior and efficient solution for solving multiphase flow problems. It outperforms traditional simulators in accuracy, speed, and data utilization.
ADVANCES IN WATER RESOURCES
(2022)
Article
Mathematics, Applied
Alexander Kovacs et al.
Summary: This paper introduces conditional PINNs for estimating the solutions of classes of eigenvalue problems. It demonstrates the ability of neural networks to learn solutions for entire classes of problems and shows applications in physics of magnetization reversal.
COMMUNICATIONS IN NONLINEAR SCIENCE AND NUMERICAL SIMULATION
(2022)
Article
Engineering, Multidisciplinary
Pao-Hsiung Chiu et al.
Summary: This study proposes a novel physics-informed neural network (PINN) method that combines automatic differentiation (AD) and numerical differentiation (ND) to improve the training efficiency and accuracy. The proposed method, called canPINN, achieves more robust and efficient training than AD-based PINNs, while improving accuracy by up to 1-2 orders of magnitude relative to ND-based PINNs. The method is demonstrated on challenging fluid dynamic problems, showing consistently high accuracy compared to conventional AD-based PINNs.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2022)
Article
Mathematics, Applied
Siddhartha Mishra et al.
Summary: This article focuses on the application of physics-informed neural networks (PINNs) to approximating inverse problems for partial differential equations (PDEs). Specifically, it investigates data assimilation or unique continuation problems and provides rigorous estimates on the generalization error of PINNs for these problems. The study establishes an abstract framework and utilizes conditional stability estimates for the inverse problem to derive the estimate on PINN generalization error, thereby justifying the use of PINNs in this context.
IMA JOURNAL OF NUMERICAL ANALYSIS
(2022)
Article
Mathematics, Applied
A. Biswas et al.
Summary: In this study, error estimates and stability analysis of deep learning techniques for certain partial differential equations, including the incompressible Navier-Stokes equations, are provided. Explicit error estimates (in suitable norms) are obtained for the solution computed by optimizing a loss function in a Deep Neural Network approximation of the solution, with a fixed complexity.
NUMERISCHE MATHEMATIK
(2022)
Article
Computer Science, Interdisciplinary Applications
Lei Yuan et al.
Summary: Physics informed neural networks (PINNs) are a novel deep learning paradigm for solving forward and inverse problems of nonlinear partial differential equations (PDEs). This study proposes an auxiliary PINN (A-PINN) framework that bypasses the limitation of integral discretization by defining auxiliary output variables and using automatic differentiation. The A-PINN demonstrates higher accuracy compared to traditional PINN and shows good performance in solving various nonlinear IDE problems.
JOURNAL OF COMPUTATIONAL PHYSICS
(2022)
Article
Computer Science, Artificial Intelligence
Zhiwei Fang
Summary: The article introduces a hybrid physics-informed neural network (hybrid PINN) for solving partial differential equations (PDEs) by using an approximation of the differential operator. This method has a convergent rate, avoiding the issue of bad predictions by neural networks. It is the first work to have a machine learning PDE solver with a convergent rate like numerical methods.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Mathematics, Applied
Tim De Ryck et al.
Summary: Physics-informed neural networks can approximate solutions of PDEs by minimizing pointwise residuals, and rigorous bounds on the error are derived in this study. The research also proves that the total error can be bounded by the generalization error, and the size of the neural networks and the number of training samples can grow polynomially with the underlying dimension.
ADVANCES IN COMPUTATIONAL MATHEMATICS
(2022)
Article
Geochemistry & Geophysics
Jonthan D. Smith et al.
Summary: This approach introduces a probabilistic hypocenter inversion method using Stein variational inference with a differentiable physics informed neural network model. It is capable of efficiently handling highly multimodal posterior distributions in hypocentral inverse problems, and is applicable to seismic network geometries without the need for traveltime tables.
GEOPHYSICAL JOURNAL INTERNATIONAL
(2022)
Review
Computer Science, Artificial Intelligence
Amina N. Muhammad et al.
Summary: The purpose of smart cities is to improve resource utilization efficiency and enhance residents' quality of life, mainly optimizing city functions through data generated by IoT devices; recent attention has been given to the application of deep learning in smart cities, but there is currently no dedicated survey to showcase its progress and future directions.
NEURAL COMPUTING & APPLICATIONS
(2021)
Article
Computer Science, Interdisciplinary Applications
Sifan Wang et al.
Summary: Free boundary problems are common in mathematics, science, and engineering, posing a computational challenge that requires accurate numerical methods. This work introduces a multi-network model based on physics-informed neural networks to address Stefan problems, both forward and inverse. The effectiveness of this approach is demonstrated across various benchmarks, showcasing the accurate recovery of solutions involving dynamic boundaries in partial differential equations.
JOURNAL OF COMPUTATIONAL PHYSICS
(2021)
Article
Jan Blechschmidt et al.
Review
Engineering, Mechanical
Shengze Cai et al.
Summary: Significant progress has been made in simulating flow problems over the last 50 years, but challenges remain in incorporating noisy data, complex mesh generation, and solving high-dimensional problems. Physics-informed neural networks (PINNs) have been demonstrated as effective in solving inverse flow problems related to various fluid dynamics scenarios.
ACTA MECHANICA SINICA
(2021)
Article
Materials Science, Multidisciplinary
Mahmudul Islam et al.
Summary: Using multi-fidelity physics informed neural networks (MPINN) to simulate long physical processes via molecular dynamics proves to be a cost-effective and accurate approach. By determining the optimal percentage of high-fidelity training data required, high system properties accuracy can be achieved while saving computational costs up to 68%. These results demonstrate the effectiveness of MPINN in predicting nanofluid viscosity accurately with significantly fewer MD simulations, opening doors to investigate more complex engineering problems requiring long-range MD simulations.
COMPUTATIONAL MATERIALS SCIENCE
(2021)
Article
Mathematics, Interdisciplinary Applications
Qiming Zhu et al.
Summary: This study investigates the application of physics-informed neural network (PINN) in metal additive manufacturing (AM) by accurately predicting temperature and melt pool dynamics during the process with a moderate amount of labeled data. The fusion of data and physical principles demonstrates the potential of PINN for broader applications in advanced manufacturing.
COMPUTATIONAL MECHANICS
(2021)
Article
Engineering, Multidisciplinary
Ehsan Haghighat et al.
Summary: SciANN is a Python package for scientific computing and physics-informed deep learning. It utilizes TensorFlow and Keras to build deep neural networks and optimization models, allowing for solving partial differential equations with flexibility in setting complex functional forms.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2021)
Article
Physics, Multidisciplinary
Luka Grubisic et al.
Summary: The study focuses on eigenmode localization for a class of elliptic reaction-diffusion operators, using a family of Schrodinger Hamiltonians as the prototype model problem to compute localized bounded states. Various deep network architectures are presented to predict the localization of bounded states from potential samples, with a focus on interpretability. A physics-informed deep dense network is employed to handle higher dimensional problems, and experiments are conducted to measure the accuracy and performance of the proposed algorithm.
Article
Engineering, Multidisciplinary
Kailai Xu et al.
Summary: Inverse problems associated with stochastic models, where the unknown quantities are distributions, often face limitations with traditional methods that require closed-form density functions or a large number of simulations. A new method is proposed in this study which utilizes neural networks to approximate the unknown distribution and compute statistical discrepancies. Numerical experiments demonstrate the effectiveness of the proposed method in estimating model parameters and learning complex unknown distributions.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2021)
Article
Engineering, Multidisciplinary
Sina Amini Niaki et al.
Summary: The study introduces a Physics-Informed Neural Network (PINN) to simulate the thermal-chemical evolution of a composite material curing in an autoclave. By optimizing deep neural network (DNN) parameters using a physics-based loss function, the research solves coupled differential equations, designs a PINN with two disconnected subnetworks, and develops a sequential training algorithm. The approach incorporates explicit discontinuities at the composite-tool interface and enforces known physical behavior in the loss function to enhance solution accuracy.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2021)
Article
Engineering, Multidisciplinary
Ehsan Haghighat et al.
Summary: The Physics-Informed Neural Network (PINN) framework combines physics with deep learning to solve PDEs and identify parameters. A nonlocal PINN approach with long-range interactions shows superior performance in solution accuracy and parameter inference compared to local approaches.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2021)
Article
Engineering, Multidisciplinary
Nanzhe Wang et al.
Summary: A Theory-guided Auto-Encoder (TgAE) framework is proposed for surrogate construction, uncertainty quantification, and inverse modeling tasks. TgAE is trained based on the Auto-Encoder architecture of CNN with discretized governing equations embedded for theory-guided training. Results demonstrate satisfactory accuracy and efficiency for surrogate modeling and uncertainty quantification tasks.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2021)
Article
Computer Science, Interdisciplinary Applications
Umair bin Waheed et al.
Summary: The proposed algorithm based on physics-informed neural networks offers a high traveltime accuracy for a wide range of applications in seismology. It leverages machine learning techniques like transfer learning and surrogate modeling to speed up computations for updated velocity models and source locations. The method's flexibility in incorporating various physical properties and improving convergence rate makes it an efficient forward modeling engine for seismological applications.
COMPUTERS & GEOSCIENCES
(2021)
Article
Computer Science, Interdisciplinary Applications
Liu Yang et al.
Summary: The paper introduces a Bayesian physics-informed neural network (B-PINN) for solving nonlinear problems, and conducts experimental comparisons on posterior estimation methods and uncertainty quantification. The KL model is as accurate and faster than BNN, but cannot be applied to high-dimensional problems.
JOURNAL OF COMPUTATIONAL PHYSICS
(2021)
Article
Geochemistry & Geophysics
Jonathan D. Smith et al.
Summary: EikoNet is a deep learning approach proposed to solve the Eikonal equation, allowing for rapid determination of travel time between any two points in a continuous 3-D domain. By exploiting the differentiability of neural networks, the method calculates spatial gradients analytically without needing solutions from a finite-difference algorithm, and has been rigorously tested on velocity models and sampling methods.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2021)
Article
Computer Science, Interdisciplinary Applications
Amuthan A. Ramabathiran et al.
Summary: The paper introduces a class of Sparse, Physics-based, and partially Interpretable Neural Networks (SPINN) for solving ordinary and partial differential equations (PDEs) by reinterpreting a traditional meshless representation of solutions of PDEs. The SPINN model serves as a bridge between dense neural network based methods and traditional mesh-free numerical methods, offering a novel means to develop hybrid algorithms. A key feature of SPINN is its interpretability and sparsity, with fewer connections than typical dense neural networks used for PDEs, along with implicit mesh adaptivity and the ability to handle solution discontinuities.
JOURNAL OF COMPUTATIONAL PHYSICS
(2021)
Article
Computer Science, Interdisciplinary Applications
Zhiping Mao et al.
Summary: This study introduces a novel efficient approach using neural networks to predict fluid properties changes in high-speed flow. By training and utilizing DeepONet, rapid and accurate predictions can be achieved, maintaining good predictive performance even when sparse measurements are available.
JOURNAL OF COMPUTATIONAL PHYSICS
(2021)
Article
Mechanics
Chensen Lin et al.
Summary: The article introduces a seamless method based on the deep neural network DeepONet for modeling the bubble growth dynamics of multiscale systems. It can accurately predict the bubble growth process across a wide range of spatial and temporal scales.
JOURNAL OF FLUID MECHANICS
(2021)
Article
Engineering, Multidisciplinary
Ehsan Kharazmi et al.
Summary: The study proposes a general framework for hp-variational physics-informed neural networks (hp-VPINNs) based on the nonlinear approximation of shallow and deep neural networks and hp-refinement via domain decomposition and projection onto high-order polynomials. The hp-refinement corresponds to a global approximation with a local learning algorithm for efficient network parameter optimization, demonstrating advantages in accuracy and training cost for function approximation and solving differential equations.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2021)
Article
Computer Science, Interdisciplinary Applications
Felipe A. C. Viana et al.
Summary: The paper proposes an approach that uses observed data to estimate missing physics in the original model, utilizing recurrent neural networks and directed graphs to quantify the missing physics and reduce the gap between predictions and observations. The study demonstrates the ability to estimate discrepancies at reasonable computational cost through three case studies and state-of-the-art machine learning software packages.
COMPUTERS & STRUCTURES
(2021)
Article
Computer Science, Interdisciplinary Applications
Yigit A. Yucesan et al.
Summary: The field failures of wind turbine main bearings can lead to undesired downtime and significant maintenance costs. A hybrid model calibrated with visual grease inspections can successfully predict bearing fatigue prognosis even with limited and noisy field observations. This model can help optimize park reliability by suggesting turbine-specific regreasing intervals.
COMPUTERS IN INDUSTRY
(2021)
Article
Water Resources
Behzad Jamali et al.
Summary: The study investigates the use of Machine Learning to improve the speed of flood modelling by proposing a new method that integrates ML with hydrodynamic models. Testing two ML approaches, curve fitting and artificial neural networks, the study finds that curve fitting method can be successfully implemented with improved generalization performance.
HYDROLOGICAL PROCESSES
(2021)
Article
Computer Science, Artificial Intelligence
Tim De Ryck et al.
Summary: In this study, bounds on the error incurred in the approximation of Sobolev-regular and analytic functions by neural networks with the hyperbolic tangent activation function were derived, providing explicit estimates on the approximation error with respect to the size of the neural networks. It was demonstrated that tanh neural networks with only two hidden layers are sufficient to approximate functions at comparable or better rates than much deeper ReLU neural networks.
Article
Computer Science, Artificial Intelligence
Enrico Schiassi et al.
Summary: X-TFC is a physics-informed neural network method that combines the Theory of Functional Connections and Neural Networks to solve differential equation problems accurately and quickly. It achieves high accuracy with low computational time, even for large scale problems, by using a single-layer neural network trained via the Extreme Learning Machine algorithm.
Article
Chemistry, Physical
Chensen Lin et al.
Summary: This study explores the use of deep operator networks (DeepONet) to learn dynamics at different scales and demonstrates their accuracy and effectiveness in predicting multirate bubble growth dynamics. Results show that properly trained DeepONets can accurately predict macroscale bubble growth dynamics and outperform long short-term memory networks with only a few new measurements.
JOURNAL OF CHEMICAL PHYSICS
(2021)
Article
Computer Science, Interdisciplinary Applications
Han Gao et al.
Summary: The paper introduces a novel physics-constrained CNN learning architecture to learn solutions of parametric PDEs on irregular domains without labeled data. Elliptic coordinate mapping is used to enable coordinate transforms between irregular physical domain and regular reference domain. The proposed method has been assessed by solving a number of steady-state PDEs on irregular domains and has shown notable superiority over the FC-NN based PINN in terms of efficiency and accuracy.
JOURNAL OF COMPUTATIONAL PHYSICS
(2021)
Article
Computer Science, Interdisciplinary Applications
Xiaowei Jin et al.
Summary: In the past 50 years, significant progress has been made in solving Navier-Stokes equations using various numerical methods, but challenges still exist in incorporating multi-fidelity data seamlessly and mesh generation for complex industrial applications. Physics-informed neural networks (PINNs) are employed to directly encode governing equations into deep neural networks, overcoming some limitations for simulating incompressible laminar and turbulent flows. The Navier-Stokes flow nets (NSFnets) are developed using velocity-pressure (VP) and vorticity-velocity (VV) formulations, showing promise in accuracy and convergence rate for benchmark problems as well as turbulent channel flows.
JOURNAL OF COMPUTATIONAL PHYSICS
(2021)
Review
Multidisciplinary Sciences
K. Kashinath et al.
Summary: This study explores systematic approaches to incorporating physics and domain knowledge into machine learning models, showing successful applications in emulating, downscaling, and forecasting weather and climate processes. Through 10 case studies, it demonstrates improvements in physical consistency, reduced training time, enhanced data efficiency, and better generalization.
PHILOSOPHICAL TRANSACTIONS OF THE ROYAL SOCIETY A-MATHEMATICAL PHYSICAL AND ENGINEERING SCIENCES
(2021)
Article
Environmental Sciences
Chen Cheng et al.
Summary: This study proposes a method using a Physical Informed Neural Network (PINN) combined with Resnet blocks to solve fluid flow problems. The results show that the Res-PINN outperforms traditional deep learning methods in predictive ability and is capable of predicting velocity and pressure fields in spatial-temporal fluid flows. The errors in inverse parameters for fluid flow problems are also well-controlled, demonstrating the effectiveness of the Res-PINN approach.
Article
Multidisciplinary Sciences
Henry Cavanagh et al.
Summary: The study introduces a novel framework to characterize morphodynamics for cell-drug interactions directly from images and explains the perturbed development of the Asian soybean rust pathogen. Various landscapes describing phenotype transitions during growth have been discovered, identifying possible perturbations in the tip growth machinery causing variation. This demonstrates a widely-applicable integration of unsupervised learning and biophysical modeling.
NATURE COMMUNICATIONS
(2021)
Review
Optics
Peter R. Wiecha et al.
Summary: Deep learning in nano-photonics is mostly discussed in terms of its potential for inverse design of photonic devices or nano-structures. While specific, drawbacks of the approaches require critical evaluation. Additionally, deep learning has evolved as a tool for a variety of applications such as accelerating photonics simulations, data reconstruction, imaging, and knowledge discovery.
PHOTONICS RESEARCH
(2021)
Article
Computer Science, Interdisciplinary Applications
Mohammad Amin Nabian et al.
Summary: PINNs are deep neural networks trained to compute the response of systems governed by PDEs using automatic differentiation. Although successful, they still need improvements in computational efficiency, which is why this paper studies the performance of an importance sampling approach for efficient training of PINNs.
COMPUTER-AIDED CIVIL AND INFRASTRUCTURE ENGINEERING
(2021)
Article
Automation & Control Systems
Florian Arnold et al.
Summary: This paper introduces a state-space model formulation using physics-informed neural networks to efficiently model complex dynamic systems. The resulting model does not require numerical solution techniques during state propagation and is suitable for real-time applications.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2021)
Article
Chemistry, Multidisciplinary
Johannes G. Hoffer et al.
Summary: This paper discusses and compares several classes of mesh-free surrogate models based on traditional and thriving machine learning and deep learning methods, showing that relatively simple algorithms are competitive in applications with low geometric complexity and extrapolation requirements, while complex DL methods require further research before practical use.
APPLIED SCIENCES-BASEL
(2021)
Article
Computer Science, Artificial Intelligence
Jian Cheng Wong et al.
Summary: This paper introduces the application of neuroevolution in solving differential equations and proposes a novel transfer neuroevolution algorithm. The algorithm demonstrates improved efficiency in solving optimization problems, resulting in higher accuracy and faster convergence. Experimental results show that this method outperforms traditional stochastic gradient descent (SGD) in solving differential equations.
IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE
(2021)
Article
Computer Science, Information Systems
Dennis Elbrachter et al.
Summary: This study explores the fundamental limits of deep neural network learning, revealing the Kolmogorov-optimal approximation and exponential approximation accuracy in a variety of function classes. Results show that deep networks require strictly smaller connectivity than wide networks in the approximation of sufficiently smooth functions.
IEEE TRANSACTIONS ON INFORMATION THEORY
(2021)
Article
Thermodynamics
Shengze Cai et al.
Summary: Physics-informed neural networks (PINNs) have gained popularity in engineering fields for their effectiveness in solving realistic problems with noisy data and partially missing physics. Through automatic differentiation to evaluate differential operators and defining a multitask learning problem, PINNs have been applied to various heat transfer problems, bridging the gap between computational and experimental heat transfer.
JOURNAL OF HEAT TRANSFER-TRANSACTIONS OF THE ASME
(2021)
Review
Engineering, Mechanical
Sung Wook Kim et al.
Summary: Despite the rapid development of AI, limitations such as lack of robustness and interpretability have hindered its widespread adoption. To overcome these limitations, new branches of deep learning such as physics-informed neural networks have emerged, providing new opportunities for advancement in the field.
JOURNAL OF MECHANICAL SCIENCE AND TECHNOLOGY
(2021)
Article
Acoustics
Zhilu Lai et al.
Summary: This paper introduces a new approach of using Physics-informed Neural ODEs for structural identification, which enhances adaptability and flexibility through a two-level representation involving physics information and discrepancy terms, addressing complex issues in structural identification.
JOURNAL OF SOUND AND VIBRATION
(2021)
Article
Computer Science, Artificial Intelligence
Dehao Liu et al.
Summary: A new physics-constrained neural network with the minimax architecture (PCNN-MM) is proposed in this paper, which allows for systematic adjustment of weights of different losses. A novel saddle point search algorithm called Dual-Dimer method is developed, showing computational efficiency and providing additional eigenvalue information. The convergence of PCNN-MMs is demonstrated to be faster than traditional PCNNs, especially for nonconvex-nonconcave functions.
Article
Engineering, Electrical & Electronic
Bernhard Schoelkopf et al.
Summary: The fields of machine learning and graphical causality have started to influence each other and show interest in benefiting from each other's advancements. Understanding fundamental concepts of causal inference, and relating them to key issues in machine learning, can help enhance modern machine learning research. A central problem in the intersection of AI and causality is the learning of causal representations, which involves discovering high-level causal variables from low-level observations.
PROCEEDINGS OF THE IEEE
(2021)
Article
Engineering, Multidisciplinary
Ehsan Haghighat et al.
Summary: This study presents the application of Physics Informed Neural Networks (PINN) in solid mechanics, improving accuracy and convergence with a multi-network model and Isogeometric Analysis. The study demonstrates the importance of honoring physics in improving robustness and highlights the potential application of PINN in sensitivity analysis and surrogate modeling.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2021)
Article
Engineering, Multidisciplinary
Wei Li et al.
Summary: This study establishes a neural network-based computational framework to characterize the finite deformation of elastic plates by incorporating known physical laws into the training process, reducing the data demand significantly. The accuracy of the modeling framework is carefully examined by applying it to different loading cases, showing satisfactory results with proper training strategies.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2021)
Article
Computer Science, Interdisciplinary Applications
Christopher J. Arthurs et al.
Summary: The goal of this work is to train a neural network that approximates solutions to the Navier-Stokes equations, with three main contributions: demonstrating the efficiency of neural networks in aggregating parametric solutions, interpolating finite element solutions, and introducing an active learning algorithm. Additionally, the network has the capability of accurately predicting narrowing in a tube for medical diagnosis and computer-aided design applications during rapid parameter sweeping.
JOURNAL OF COMPUTATIONAL PHYSICS
(2021)
Article
Computer Science, Interdisciplinary Applications
Yunjin Tong et al.
Summary: The researchers proposed a novel neural network architecture Taylor-nets for long-term prediction of Hamiltonian dynamic systems while preserving their symplectic structures. The algorithm performs well in terms of prediction accuracy, convergence rate, and robustness, even with extremely small training data and a short training period.
JOURNAL OF COMPUTATIONAL PHYSICS
(2021)
Article
Computer Science, Interdisciplinary Applications
Shengze Cai et al.
Summary: This article introduces a new data assimilation framework, DeepM & Mnet, for simulating multiphysics and multiscale problems using pre-trained neural networks. The framework utilizes DeepONets to quickly and accurately build complex models based on very few measurements.
JOURNAL OF COMPUTATIONAL PHYSICS
(2021)
Article
Optics
Siddhartha Mishra et al.
Summary: The study introduces a machine learning algorithm based on physics informed neural networks for simulating and solving radiative transfer problems, demonstrating superior accuracy and efficiency through empirical experiments.
JOURNAL OF QUANTITATIVE SPECTROSCOPY & RADIATIVE TRANSFER
(2021)
Article
Physics, Multidisciplinary
Li Wang et al.
Summary: This paper uses the multi-layer PINN deep learning method to study data-driven rogue wave solutions of the defocusing nonlinear Schrodinger (NLS) equation with time-dependent potential. By considering various initial conditions and periodic boundary conditions, the study discusses learning parameters in the context of rogue wave solutions.
Proceedings Paper
Green & Sustainable Science & Technology
Jochen Stiasny et al.
Summary: This paper investigates the performance of PINNs for discovering the frequency dynamics of future power systems, aiming to address challenges such as stronger non-linearities of low-inertia systems, increased measurement noise, and limited availability of data.
2021 IEEE MADRID POWERTECH
(2021)
Article
Computer Science, Artificial Intelligence
Ali Girayhan Ozbay et al.
Summary: The study introduces a novel fully convolutional neural network architecture for solving the Poisson equation, providing flexibility in handling partial differential equations. The model is able to address boundary conditions using a unique approach and is trained with a novel loss function. Results show that the model performs well in solving the Poisson equation, aiding in improving computational efficiency.
DATA-CENTRIC ENGINEERING
(2021)
Article
Mathematics, Applied
Lu Lu et al.
Summary: Inverse design, such as topology optimization, is widely used in engineering for achieving targeted properties by optimizing designed geometries. The proposed physics-informed neural networks with hard constraints (hPINNs) can effectively solve topology optimization problems without the need for a large dataset, demonstrating smoother design outcomes compared to conventional methods.
SIAM JOURNAL ON SCIENTIFIC COMPUTING
(2021)
Article
Computer Science, Information Systems
Stefano Markidis
Summary: The study explores the potential of Physics-Informed Neural Networks (PINN) as linear solvers for solving partial differential equations, particularly focusing on the Poisson equation. It is found that low-frequency components of the solution converge quickly, while accurate solutions for high frequencies require a significantly longer time. The integration of PINNs into traditional linear solvers leads to the development of new solvers with performance comparable to high-performance solvers such as PETSc conjugate gradient linear solvers.
FRONTIERS IN BIG DATA
(2021)
Article
Mathematics, Applied
Sifan Wang et al.
Summary: The study reviews recent advances in scientific machine learning, focusing specifically on the effectiveness of physics-informed neural networks in predicting outcomes of physical systems and discovering hidden physics from noisy data. It also proposes a learning rate annealing algorithm and a novel neural network architecture to address numerical stiffness issues in training constrained neural networks.
SIAM JOURNAL ON SCIENTIFIC COMPUTING
(2021)
Article
Computer Science, Interdisciplinary Applications
Ehsan Kharazmi et al.
Summary: This study employs physics-informed neural networks to analyze various epidemiological models, identifying time-dependent parameters and data-driven fractional differential operators, while inferring the spread of COVID-19 in different regions. The results show that identifying time-dependent parameters using neural networks is suitable for integer-order and time-delay models, while determining different time-dependent derivative orders using neural networks is suitable for fractional differential models.
NATURE COMPUTATIONAL SCIENCE
(2021)
Article
Computer Science, Hardware & Architecture
Ka Chun Cheung et al.
Summary: Machine learning methods have been successfully applied in various fields, including scientific computing, to solve complex problems such as partial differential equations. This data-driven approach has shown promising results in improving computational efficiency and model accuracy.
CCF TRANSACTIONS ON HIGH PERFORMANCE COMPUTING
(2021)
Article
Computer Science, Artificial Intelligence
Christopher Irrgang et al.
Summary: In recent years, artificial intelligence (AI) methods have been increasingly used to enhance Earth and climate modeling. This Perspective examines the opportunity to go further and build hybrid systems that integrate AI tools and models based on physical process knowledge to make more efficient use of daily observational data streams.
NATURE MACHINE INTELLIGENCE
(2021)
Article
Physics, Fluids & Plasmas
Thomas Stielow et al.
Summary: Single-shot wide-angle diffraction imaging is a method widely used for investigating the structure of noncrystallizing objects, with no need for tomographic measurements to reconstruct the object's three-dimensional structure. Neural networks excel in image processing tasks and can be utilized for reconstructing object structures.
Review
Physics, Applied
George Em Karniadakis et al.
Summary: Physics-informed learning seamlessly integrates data and mathematical models through neural networks or kernel-based regression networks for accurate inference of realistic and high-dimensional multiphysics problems. Challenges remain in incorporating noisy data seamlessly, complex mesh generation, and addressing high-dimensional problems.
NATURE REVIEWS PHYSICS
(2021)
Article
Computer Science, Artificial Intelligence
Lu Lu et al.
Summary: This study extends the capabilities of neural networks with the introduction of the deep operator network (DeepONet), which can be used to learn various operators, including explicit and implicit operators. Different formulations of the input function space were studied and their effect on generalization error for 16 diverse applications was examined.
NATURE MACHINE INTELLIGENCE
(2021)
Article
Mathematics, Applied
Lu Lu et al.
Summary: This article introduces an overview, implementation and applications of Physics-Informed Neural Networks (PINNs), along with a new residual-based adaptive refinement (RAR) method. By comparing with finite element methods, the advantages of PINNs and the versatile applications of the Python library DeepXDE are demonstrated. Overall, DeepXDE contributes to the more rapid development of the emerging scientific machine learning field.
Proceedings Paper
Computer Science, Hardware & Architecture
Aditya Balu et al.
Summary: The study focuses on the distributed training of large scale neural networks as PDE solvers, introducing a scalable framework that integrates multigrid technique and distributed deep learning framework to significantly reduce solve time and predict output solutions at high resolutions.
SC21: INTERNATIONAL CONFERENCE FOR HIGH PERFORMANCE COMPUTING, NETWORKING, STORAGE AND ANALYSIS
(2021)
Proceedings Paper
Acoustics
Shaikhah Alkhadhr et al.
Summary: The article discusses the importance of simulating the wave equation in medical ultrasound applications, highlights the potential of deep neural networks for solving PDEs, and presents the use of PINNs to simulate the numerical solution of the wave equation.
INTERNATIONAL ULTRASONICS SYMPOSIUM (IEEE IUS 2021)
(2021)
Review
Computer Science, Interdisciplinary Applications
Shaveta Dargan et al.
ARCHIVES OF COMPUTATIONAL METHODS IN ENGINEERING
(2020)
Article
Engineering, Multidisciplinary
Luning Sun et al.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2020)
Article
Computer Science, Interdisciplinary Applications
Ameya D. Jagtap et al.
JOURNAL OF COMPUTATIONAL PHYSICS
(2020)
Article
Computer Science, Interdisciplinary Applications
Nicholas Geneva et al.
JOURNAL OF COMPUTATIONAL PHYSICS
(2020)
Article
Computer Science, Artificial Intelligence
Arwa Aldweesh et al.
KNOWLEDGE-BASED SYSTEMS
(2020)
Article
Engineering, Multidisciplinary
Georgios Kissas et al.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2020)
Article
Engineering, Multidisciplinary
Zhiping Mao et al.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2020)
Article
Computer Science, Interdisciplinary Applications
Heng Xiao et al.
COMPUTERS & FLUIDS
(2020)
Review
Computer Science, Artificial Intelligence
Saptarshi Sengupta et al.
KNOWLEDGE-BASED SYSTEMS
(2020)
Article
Multidisciplinary Sciences
Maziar Raissi et al.
Article
Mathematics, Applied
Liu Yang et al.
SIAM JOURNAL ON SCIENTIFIC COMPUTING
(2020)
Article
Physics, Multidisciplinary
Francisco Sahli Costabal et al.
FRONTIERS IN PHYSICS
(2020)
Article
Multidisciplinary Sciences
Lu Lu et al.
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
(2020)
Article
Engineering, Mechanical
Somdatta Goswami et al.
THEORETICAL AND APPLIED FRACTURE MECHANICS
(2020)
Article
Environmental Sciences
A. M. Tartakovsky et al.
WATER RESOURCES RESEARCH
(2020)
Article
Optics
Yuyao Chen et al.
Article
Water Resources
QiZhi He et al.
ADVANCES IN WATER RESOURCES
(2020)
Article
Materials Science, Multidisciplinary
M. Torabi Rad et al.
COMPUTATIONAL MATERIALS SCIENCE
(2020)
Article
Engineering, Multidisciplinary
Ameya D. Jagtap et al.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2020)
Article
Automation & Control Systems
Shiliang Sun et al.
IEEE TRANSACTIONS ON CYBERNETICS
(2020)
Article
Engineering, Multidisciplinary
Vien Minh Nguyen-Thanh et al.
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
(2020)
Article
Engineering, Multidisciplinary
Xuhui Meng et al.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2020)
Article
Engineering, Multidisciplinary
Ruiyang Zhang et al.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2020)
Article
Physics, Mathematical
Zhi-Qin John Xu et al.
COMMUNICATIONS IN COMPUTATIONAL PHYSICS
(2020)
Article
Physics, Mathematical
Yeonjong Shin et al.
COMMUNICATIONS IN COMPUTATIONAL PHYSICS
(2020)
Article
Automation & Control Systems
Renato G. Nascimento et al.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2020)
Article
Mathematics, Applied
Dongkun Zhang et al.
SIAM JOURNAL ON SCIENTIFIC COMPUTING
(2020)
Article
Computer Science, Information Systems
Zhiwei Fang et al.
Article
Computer Science, Information Systems
Zhiwei Fang et al.
Article
Computer Science, Interdisciplinary Applications
M. Raissi et al.
JOURNAL OF COMPUTATIONAL PHYSICS
(2019)
Article
Computer Science, Interdisciplinary Applications
Houman Owhadi et al.
JOURNAL OF COMPUTATIONAL PHYSICS
(2019)
Review
Computer Science, Information Systems
Md Zahangir Alom et al.
Article
Computer Science, Interdisciplinary Applications
Yinhao Zhu et al.
JOURNAL OF COMPUTATIONAL PHYSICS
(2019)
Article
Computer Science, Interdisciplinary Applications
Yibo Yang et al.
JOURNAL OF COMPUTATIONAL PHYSICS
(2019)
Article
Computer Science, Interdisciplinary Applications
Dongkun Zhang et al.
JOURNAL OF COMPUTATIONAL PHYSICS
(2019)
Article
Multidisciplinary Sciences
Mikhail Belkin et al.
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
(2019)
Article
Statistics & Probability
Benedikt Bauer et al.
ANNALS OF STATISTICS
(2019)
Review
Behavioral Sciences
Marta Garnelo et al.
CURRENT OPINION IN BEHAVIORAL SCIENCES
(2019)
Article
Computer Science, Artificial Intelligence
Vikas Dwivedi et al.
Article
Mathematics, Applied
Pavan Pranjivan Mehta et al.
FRACTIONAL CALCULUS AND APPLIED ANALYSIS
(2019)
Article
Mathematics, Applied
Guofei Pang et al.
SIAM JOURNAL ON SCIENTIFIC COMPUTING
(2019)
Review
Computer Science, Information Systems
Daniel S. Berman et al.
Review
Computer Science, Information Systems
Ajay Shrestha et al.
Review
Pharmacology & Pharmacy
Hongming Chen et al.
DRUG DISCOVERY TODAY
(2018)
Article
Computer Science, Interdisciplinary Applications
Maziar Raissi et al.
JOURNAL OF COMPUTATIONAL PHYSICS
(2018)
Article
Mathematics, Applied
Maziar Raissi et al.
SIAM JOURNAL ON SCIENTIFIC COMPUTING
(2018)
Article
Computer Science, Interdisciplinary Applications
Justin Sirignano et al.
JOURNAL OF COMPUTATIONAL PHYSICS
(2018)
Article
Computer Science, Artificial Intelligence
Jens Berg et al.
Article
Mathematics
E. Weinan et al.
COMMUNICATIONS IN MATHEMATICS AND STATISTICS
(2018)
Article
Engineering, Electrical & Electronic
Kai Arulkumaran et al.
IEEE SIGNAL PROCESSING MAGAZINE
(2017)
Article
Computer Science, Interdisciplinary Applications
Maziar Raissi et al.
JOURNAL OF COMPUTATIONAL PHYSICS
(2017)
Article
Computer Science, Interdisciplinary Applications
Maziar Raissi et al.
JOURNAL OF COMPUTATIONAL PHYSICS
(2017)
Article
Computer Science, Artificial Intelligence
Dmitry Yarotsky
Review
Multidisciplinary Sciences
Stephane Mallat
PHILOSOPHICAL TRANSACTIONS OF THE ROYAL SOCIETY A-MATHEMATICAL PHYSICAL AND ENGINEERING SCIENCES
(2016)
Article
Mathematics, Interdisciplinary Applications
Houman Owhadi
MULTISCALE MODELING & SIMULATION
(2015)
Review
Multidisciplinary Sciences
Yann LeCun et al.
Article
Mathematics, Applied
Manoj Kumar et al.
COMPUTERS & MATHEMATICS WITH APPLICATIONS
(2011)
Article
Computer Science, Artificial Intelligence
Guang-Bin Huang et al.
INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS
(2011)
Article
Computer Science, Artificial Intelligence
IE Lagaris et al.
IEEE TRANSACTIONS ON NEURAL NETWORKS
(2000)
