Lost in Latent Space: An Empirical Study of Latent Diffusion Models for Physics Emulation
This addresses the problem of slow inference for physics emulation, offering a practical solution with significant speed-ups, though it is incremental as it adapts an existing strategy from image generation to a new domain.
The study tackled the high computational cost of diffusion models for physics emulation by applying latent-space generation, finding that accuracy remains robust up to 1000x compression and that diffusion-based emulators outperform non-generative methods with greater diversity in predictions.
The steep computational cost of diffusion models at inference hinders their use as fast physics emulators. In the context of image and video generation, this computational drawback has been addressed by generating in the latent space of an autoencoder instead of the pixel space. In this work, we investigate whether a similar strategy can be effectively applied to the emulation of dynamical systems and at what cost. We find that the accuracy of latent-space emulation is surprisingly robust to a wide range of compression rates (up to 1000x). We also show that diffusion-based emulators are consistently more accurate than non-generative counterparts and compensate for uncertainty in their predictions with greater diversity. Finally, we cover practical design choices, spanning from architectures to optimizers, that we found critical to train latent-space emulators.