Institution: EPFL

Abstract: Researchers at EPFL have developed Dynami-CAL GraphNet, an AI algorithm designed to eliminate the "physical distortion" common in traditional AI simulations. The core breakthrough lies in embedding Newton's Third Law (action and reaction) directly into the model's architecture as a hard constraint. Unlike purely data-driven "black box" models, this algorithm ensures that all predictions adhere to the laws of conservation of momentum and energy. This advancement significantly enhances the reliability and interpretability of AI in complex dynamic systems—such as robotic gait planning and high-fidelity fluid simulations—ensuring that AI-generated results are physically viable in real-world engineering environments.

Link: https://actu.epfl.ch/news/new-ai-algorithm-is-designed-to-obey-the-laws-of-p/