Conference
Lost in Latent Space: An Empirical Study of Latent Diffusion Models for Physics Emulation
| Τίτλος: | Lost in Latent Space: An Empirical Study of Latent Diffusion Models for Physics Emulation |
|---|---|
| Συγγραφείς: | Rozet, François, Ohana, Ruben, McCabe, Michael, Louppe, Gilles, Lanusse, François, Ho, Shirley |
| Πηγή: | Advances in Neural Information Processing Systems, 38 (2025-07-03); The Thirty-Ninth Annual Conference on Neural Information Processing Systems, San Diego, United States - California [US-CA], December 2-7, 2025 |
| Στοιχεία εκδότη: | Curran Associates, 2025. |
| Έτος έκδοσης: | 2025 |
| Θεματικοί όροι: | Computer Science - Learning, Physics - Fluid Dynamics, Engineering, computing & technology, Computer science, Ingénierie, informatique & technologie, Sciences informatiques |
| Περιγραφή: | 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. |
| Τύπος εγγράφου: | conference paper http://purl.org/coar/resource_type/c_5794 conferenceObject peer reviewed |
| Γλώσσα: | English |
| Relation: | https://openreview.net/forum?id=xoNrbfbekM; https://github.com/PolymathicAI/the_well; urn:issn:1049-5258 |
| Σύνδεσμος πρόσβασης: | https://orbi.uliege.be/handle/2268/337977 |
| Rights: | open access http://purl.org/coar/access_right/c_abf2 info:eu-repo/semantics/openAccess |
| Αριθμός Καταχώρησης: | edsorb.337977 |
| Βάση Δεδομένων: | ORBi |
| Η περιγραφή δεν είναι διαθέσιμη |