Our experiments also show that latent diffusion models are consistently more accurate than deterministic neural solvers while producing diverse, statistically plausible trajectories.

When we started this project, we expected accuracy to degrade as the compression rate increased. To our surprise, we found that accuracy remained constant or even improved!

Our methodology is quite simple: we train auto-encoders to compress the state of dynamical systems (fluids, stars, ...) and train a latent diffusion model to emulate the dynamics in the compressed space. We then study the impact of the compression rate on the emulation accuracy.

Does a smaller latent space lead to worse generation in latent diffusion models? Not necessarily! We show that LDMs are extremely robust to a wide range of compression rates (10-1000x) in the context of physics emulation.

We got lost in latent space. Join us 👇

🔥 Azula 0.3.0 is out with many new features, including new super fast samplers (LMS, DPM-Solver), new guidance methods (DPS, MMPS) and new plugins (VDM, EMD, EMD2). Check it out!

I'll be in Vancouver from Dec 9 to Dec 16 to present our work at NeurIPS 2024 🇨🇦 If you want to talk about generative models, ML for science, open-source software, or bouldering, send me a DM, a comment or an email!

For short presentations, I usually introduce diffusion models in 3 slides: intuition, formalization, training/objective.

i just love @typst.app