physics-informed neural networks
PulseAugur coverage of physics-informed neural networks — every cluster mentioning physics-informed neural networks across labs, papers, and developer communities, ranked by signal.
8 天有情绪数据
-
New optimization technique boosts accuracy for complex physics neural networks
Researchers have developed a new optimization technique called SOAP+GN to improve the accuracy of physics-informed neural networks (PINNs) when dealing with complex, coupled multiphysics systems. This method addresses a…
-
Deep Neural Networks viewed as Discrete Dynamical Systems
A new research paper proposes viewing deep neural networks (DNNs) as discrete dynamical systems, drawing parallels to neural integral equations and their PDE forms. The study compares numerical solutions of Burgers' and…
-
Neural Compiler translates programs to differentiable PyTorch modules
Researchers have developed "The Neural Compiler," a system that translates symbolic programs into differentiable PyTorch modules for scientific machine learning. This approach allows for the exact encoding of known phys…
-
Hybrid physics-informed neural networks advance electricity system design
A new review paper explores the use of hybrid physics-informed neural networks (PIML) for enhancing electricity systems. These methods embed physical laws into machine learning models, improving accuracy and efficiency,…
-
StableGrad stabilizes deep neural network training without batch normalization
Researchers have introduced StableGrad, a novel optimizer-level mechanism designed to control the scale of activations and gradients in deep neural networks. This method aims to prevent training instability without rely…
-
New CGMPINN method enhances physics-informed neural network training
Researchers have developed a new method called the Curriculum-Guided Gaussian Mixture Physics-Informed Neural Network (CGMPINN) to improve the training of physics-informed neural networks (PINNs). This approach integrat…
-
New theory explains adversarial training benefits for physics-informed neural networks
Researchers have developed a new theoretical framework to understand why adversarial training improves physics-informed neural networks (PINNs). This framework, based on the influence of a GAN's discriminator on PINN tr…
-
New method boosts PDE pre-training with adaptive operator transformation
Researchers have developed AOT-POT, a novel method for pre-training neural operators on diverse partial differential equation (PDE) datasets. This approach transforms complex solution operators into simpler, aligned for…
-
New framework unifies generalization analysis for physics-informed neural networks
Researchers have developed a unified framework for analyzing the generalization capabilities of Physics-Informed Neural Networks (PINNs) and their variational counterparts (VPINNs). This new approach relaxes previous re…
-
New framework unifies generalization analysis for physics-informed neural networks
Researchers have developed a unified framework for analyzing the generalization capabilities of Physics-Informed Neural Networks (PINNs). This new approach relaxes previous restrictive assumptions and uses Taylor expans…
-
New framework tackles gradient conflict in physics-informed neural networks
Researchers have developed a new framework to address gradient conflict in physics-informed neural networks (PINNs). The approach identifies distinct conflict regimes and suggests tailored interventions, moving beyond o…
-
New method simplifies PINN training with Chebyshev center optimization
Researchers have developed a novel method for training physics-informed neural networks (PINNs) by formulating the update-direction selection as a Chebyshev-center problem. This approach aims to simplify the simultaneou…
-
New ADD-PINN framework improves traffic estimation with sparse sensor data
Researchers have developed a new framework called Adaptive Domain Decomposition Physics-Informed Neural Networks (ADD-PINN) to improve traffic state estimation from limited sensor data. This method addresses the tendenc…
-
INEUS neural solver tackles high-dimensional PIDEs with iterative regression
Researchers have developed INEUS, a novel meshfree iterative neural solver designed to tackle high-dimensional partial integro-differential equations (PIDEs). This method enhances efficiency by employing single-jump sam…
-
New framework integrates functional priors into Bayesian PINN inversion
Researchers have developed a new framework called fpBPINN to integrate functional priors into Bayesian inversion problems solved with physics-informed neural networks (PINNs). This framework addresses the challenge of d…
-
PINNs with Differentiable Chemistry Solve Stiff Reaction Systems
Researchers have developed a novel framework integrating a differentiable chemistry solver with physics-informed neural networks (PINNs) to tackle stiff and parameterized reaction systems. This approach addresses limita…
-
Adaptive Spectral PINNs show gradient scaling effects in stiff ODEs
Researchers have investigated the impact of gradient scaling in adaptive spectral Physics-Informed Neural Networks (PINNs) when applied to stiff nonlinear ordinary differential equations (ODEs). Their findings indicate …
-
CNNs on manifolds tackle boundary value problems with improved accuracy
Researchers have developed novel convolutional neural network (CNN) methods for approximating functions and solving elliptic boundary value problems on compact Riemannian manifolds. These methods demonstrate improved ap…
-
Researchers advance Physics-Informed Neural Networks for complex scientific modeling
Researchers have developed novel physics-informed neural networks (PINNs) to tackle complex differential equations. One approach, Pseudo-differential-enhanced PINNs, utilizes Fourier transforms for faster and more effic…
-
Chebyshev-Augmented OTL enables one-shot transfer learning for nonlinear PINNs
Researchers have developed a novel method called Chebyshev-Augmented One-Shot Transfer Learning (OTL) to improve the efficiency of Physics-Informed Neural Networks (PINNs). This technique addresses the limitation of PIN…