Scaling one-step generative models
Background: Generative surrogate models for dynamical systems often rely on either multi-step inference, such as diffusion-based denoising, or latent-space compression. 1-NFE (Number of Function Evaluation) models attempt to combine generative modeling with one-step rollout efficiency.
Question / Future Work: It remains an open question how generative surrogate models can be scaled to fully three-dimensional systems and multi-scale forecasting settings without sacrificing their one-step inference efficiency. Existing approaches, while successful in 2D benchmarks, require further validation and adaptation to handle the increased dimensionality and complexity of fully 3D physical dynamics.
Why It Matters: Scalability to 3D systems is the primary bottleneck for widespread adoption of generative surrogates in real-world fluid dynamics and engineering applications.
Metadata & Links
- created_at
- 2026-05-10T06:04:26Z