tum-pbs/ConFIG

[ICLR2025 Spotlight] Official implementation of Conflict-Free Inverse Gradients Method

43
/ 100
Emerging

This helps researchers and engineers who train neural networks with multiple competing objectives, such as in physics-informed neural networks or multi-task learning. It takes individual loss functions and their gradients as input and provides a single, conflict-free update direction, leading to more stable and effective model training. Scientists and engineers working on complex simulations or models will find this useful.

104 stars.

Use this if you are training a neural network where different loss terms, like physics equations and boundary conditions, conflict and prevent your model from converging effectively.

Not ideal if your neural network has a single loss function or if its multiple loss terms are already well-aligned and do not exhibit conflict.

physics-informed machine learning multi-task learning neural network training computational physics scientific machine learning
No Package No Dependents
Maintenance 6 / 25
Adoption 9 / 25
Maturity 16 / 25
Community 12 / 25

How are scores calculated?

Stars

104

Forks

10

Language

Python

License

MIT

Last pushed

Nov 17, 2025

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/tum-pbs/ConFIG"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.