How bsshreesha/physics_informed_neural_network works
Project type and purpose
• This repository is a collection of Jupyter notebooks demonstrating Physics-Informed Neural Networks (PINNs) implemented in TensorFlow to solve various partial differential equations (PDEs). It is not a traditional application with “frontend” and “backend” layers but rather an experimental ML research codebase focused on numerical PDE solutions.
File structure
• Top level contains only Jupyter notebooks (Integral_PINN.ipynb, MHD_Realistic.ipynb, etc.), one Word document (Equations.docx) and the README.
• Each notebook encapsulates a standalone experiment: data generation (domain sampling), model definition, training routine, and visualization.
• There are no separate “src” or “lib” directories—code and results live together in notebooks.
README for additional insights
• The README describes the core PINN architecture: an input layer (time/spatial coordinates), fully connected hidden layers, and an output layer predicting the PDE solution.
• The training loss combines data‐driven loss (if any measured data exist) and physics loss (enforcing the PDE residual).
• The technology stack is TensorFlow for model construction and training, and Jupyter for orchestration and visualization.