PhiFlow  by tum-pbs

Differentiable PDE solving framework for ML research

created 5 years ago
1,642 stars

Top 26.2% on sourcepulse

GitHubView on GitHub
Project Summary

Φ Flow is a Python-based simulation toolkit designed for machine learning applications, enabling end-to-end differentiable physics simulations. It targets researchers and engineers who need to integrate complex physical phenomena, particularly fluid dynamics, into deep learning models for tasks like optimization and control. The primary benefit is the seamless combination of physics-based simulations with ML frameworks, allowing for gradient-based optimization of simulation parameters or learning from simulation data.

How It Works

Φ Flow leverages automatic differentiation from popular ML frameworks (PyTorch, Jax, TensorFlow) to make its PDE solvers differentiable. This allows gradients to flow through the simulation, enabling gradient-based optimization of simulation parameters or the integration of simulations into larger neural network architectures. Its object-oriented, vectorized design and named/typed dimensions facilitate expressive, reusable, and backend-agnostic code for various dimensionalities and simulation types.

Quick Start & Requirements

Health Check
Last commit

1 day ago

Responsiveness

Inactive

Pull Requests (30d)
0
Issues (30d)
2
Star History
53 stars in the last 90 days

Explore Similar Projects

Starred by Peter Norvig Peter Norvig(Author of Artificial Intelligence: A Modern Approach; Research Director at Google), Aravind Srinivas Aravind Srinivas(Cofounder of Perplexity), and
45 more.

tensorflow by tensorflow

0.1%
191k
Open-source ML framework
created 9 years ago
updated 18 hours ago
Feedback? Help us improve.