I am Xiang CHEN, currently an AI research engineer, interested in AI for Scientific Computation. More specifically speaking:

• AI-Based Novel Solvers
  • PDE: PINN, Deep Ritz, etc.
  • Ab-Initio Simulation: FermiNet, PauliNet, DeePWF, etc.
• AI-Based Surrogate Solvers
  • PDE: FNO, DeepONet, MeshGraphNet, CFD-GCN, MP-PDE, etc.
  • Ab-Initio Simulation: Deep potential, etc.
• AI-Assisted Traditional Solvers:
  • Meshing: Meshingnet, MeshGraphNet, M2N, etc.
  • Scheme/order/stepsize control: CINN, etc.
  • Multigrid / preconditioner.
• Physics-Embedded Models:
  • Shape prior embedding (monotonicity, complexity, etc.).
  • Equivariance/invariance embedding: deepset, set trasnformer, E3NN, SchNet, etc.
  • Formulae form embedding: DeLan, LNN, Hamiltonian NN, etc.
  • Differentiable physics models
    • ODE/PDE: NeuralODE, phi-flow, jax-cfd, SU2, dolfin-adjoint, etc.
    • Optimization: OptNet, etc.
    • Fixed-Point Iteration: DEQ, etc.
    • Computational Physics/Chemistry/Material: DQC, jax-md, etc.
    • Ray-Tracing.
    • Rigid-Body Dynamics: Brax, etc.
• Inverse Design:
  • Neural reparameterization techniques.
  • Specifically designed optimization techniques.
• Applications in CEM / TCAD / mesh generation and adaptation / dynamic systems.

A literature survey on AI4PDE in Chinese is maintained at zhihu.com.

A literature survey on Physics-Embedded Machine Learning in Chinese is maintained at zhihu.com.

Check my Google Scholar Homepage for our work:

• M2N: Mesh Movement Networks for PDE Solvers
  • To the best of our knowledge the first learning-based end-to-end mesh movement framework that can greatly accelerate the mesh adaptation process by 3 to 4 orders of magnitude, whilst achieving comparable numerical error reduction to traditional sota methods.
• AD-NEGF: An End-to-End Differentiable Quantum Transport Simulator for Sensitivity Analysis and Inverse Problems
  • To the best of our knowledge the first end-to-end differentiable quantum transport simulator, which enables accurate and efficient calculation of differential physical quantities and solving inverse problems intractable by traditional optimization methods.
• SCGLED: Self-Consistent Gradient-Like Eigen Decomposition in Solving Schrodinger Equations
• BINet: learning to solve partial differential equations with boundary integral networks
• BI-GreenNet: Learning Green's functions by boundary integral network
• Performance-Guaranteed ODE Solvers with Complexity-Informed Neural Networks