Researchers have developed a novel hybrid numerical method that merges the time finite element method with deep neural networks to solve time-dependent partial differential equations. This approach differs from conventional deep learning methods by using finite element basis functions in the time direction, with a neural network defining space-dependent coefficients. The method aims to avoid statistical errors in time integration and can yield reduced spatial basis functions. An adaptive sampling strategy, employing an explicit density model to approximate PDE residual distribution, has also been developed to handle high dimensionality and improve training efficiency, as demonstrated by numerical experiments. AI
IMPACT This hybrid method could offer more efficient and accurate solutions for complex time-dependent problems in scientific computing.
RANK_REASON The cluster contains an academic paper detailing a new numerical method for solving partial differential equations. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- Deep Neural Networks
- FEM-PINN
- partial differential equations
- physics-informed neural networks
- time finite element method
- Xiaodong Feng
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
