The field of physics-informed machine learning is rapidly advancing, with a focus on developing innovative methods for modeling complex systems. Recent research has highlighted the importance of incorporating physical constraints and domain knowledge into machine learning models to improve their accuracy and reliability. One notable trend is the use of hybrid approaches that combine machine learning with traditional physical modeling techniques, such as finite element methods and partial differential equations. These approaches have shown significant promise in applications such as weather forecasting, fluid dynamics, and materials science. Noteworthy papers in this area include the proposal of DART, a framework for transforming coarse atmospheric forecasts into high-resolution satellite brightness temperature fields, and the development of SRaFTE, a two-phase learning framework for super-resolving and forecasting fine grid dynamics for time-dependent partial differential equations. Overall, the field is moving towards the development of more robust, accurate, and efficient models that can handle complex, high-dimensional data and provide insights into the underlying physical phenomena.
Advances in Physics-Informed Machine Learning for Complex Systems
Sources
Breaking the Statistical Similarity Trap in Extreme Convection Detection
Fused Lasso Improves Accuracy of Co-occurrence Network Inference in Grouped Samples
Kriging prior Regression: A Case for Kriging-Based Spatial Features with TabPFN in Soil Mapping
Conditioning on PDE Parameters to Generalise Deep Learning Emulation of Stochastic and Chaotic Dynamics
LoFT: Parameter-Efficient Fine-Tuning for Long-tailed Semi-Supervised Learning in Open-World Scenarios
Augment to Segment: Tackling Pixel-Level Imbalance in Wheat Disease and Pest Segmentation
FLARE-SSM: Deep State Space Models with Influence-Balanced Loss for 72-Hour Solar Flare Prediction
P3D: Scalable Neural Surrogates for High-Resolution 3D Physics Simulations with Global Context
Structure-Preserving High-Order Methods for the Compressible Euler Equations in Potential Temperature Formulation for Atmospheric Flows
Physics-informed sensor coverage through structure preserving machine learning
Vendi Information Gain for Active Learning and its Application to Ecology
SRaFTE: Super-Resolution and Future Time Extrapolation for Time-Dependent PDEs
Uncertainty-Aware Hourly Air Temperature Mapping at 2 km Resolution via Physics-Guided Deep Learning
Spatio-temporal DeepKriging in PyTorch: A Supplementary Application to Precipitation Data for Interpolation and Probabilistic Forecasting
Curriculum Learning for Mesh-based simulations
FOSSIL: Regret-minimizing weighting for robust learning under imbalance and small data
Learning to Retrieve for Environmental Knowledge Discovery: An Augmentation-Adaptive Self-Supervised Learning Framework
Lagrangian-Eulerian Multiscale Data Assimilation in Physical Domain based on Conditional Gaussian Nonlinear System
FlowCast-ODE: Continuous Hourly Weather Forecasting with Dynamic Flow Matching and ODE Integration
Super-Linear: A Lightweight Pretrained Mixture of Linear Experts for Time Series Forecasting