PINNide

PINNide is an open-source software framework and integrated development environment designed to streamline the creation and training of physics-informed neural networks (PINNs). It provides a high-level API to express partial differential equations, boundary conditions, and observational data, and to train models using automatic differentiation across CPU and GPU hardware.

Key features include a symbolic equation editor, a data and boundary-condition manager, collocation-point sampling strategies, and support for multiple backends, including PyTorch, TensorFlow, and JAX. PINNide offers a modular architecture with a plug-in system for solvers, optimizers, and custom loss terms, as well as built-in visualization and experiment tracking.

The architecture comprises three layers: a user-facing frontend and API, a PINN compiler that translates equations into differentiable loss components, and a training engine that orchestrates optimization, batching, and distributed computation. The project emphasizes portability, reproducibility, and interoperability with common scientific computing tools.

Development and status: PINNide originated as a research prototype in the early 2020s and has evolved through community contributions. It is maintained as an open-source project with ongoing updates to backends, solvers, and utilities.

Applications include forward and inverse PDE solving, parameter estimation, data assimilation, and surrogate modeling in engineering, physics, and applied mathematics. PINNide is used in academic studies to explore PINN architectures and in industry for rapid prototyping of physics-informed models. See also: physics-informed neural networks.