Concepts de base
Optimizing complex systems through compositional generative inverse design.
Résumé
The content introduces a novel paradigm, Compositional Generative Inverse Design (CinDM), for optimizing complex systems through inverse design. It addresses challenges in traditional optimization methods and showcases the effectiveness of diffusion models in achieving compositional generalization. The method is demonstrated in N-body interactions and 2D multi-airfoil design tasks, highlighting its ability to design systems more complex than those seen in training.
Stats
Published as a conference paper at ICLR 2024
Inverse design is crucial in various engineering domains
Recent works leverage optimization across learned dynamics models
Diffusion models improve design performance by avoiding adversarial examples
Compositional design enables combining multiple diffusion models for complex systems
Citations
"Inverse design, where we seek to design input variables in order to optimize an underlying objective function, is an important problem that arises across fields such as mechanical engineering to aerospace engineering."
"Our method generalizes to more objects for N-body dataset and discovers formation flying to minimize drag in the multi-airfoil design task."