mlsanityData to 7 Oct 2026

Lost in Latent Space: An Empirical Study of Latent Diffusion Models for Physics Emulation

arXiv 2507.02608 ↗First posted 3 Jul 2025cs.LG physics.flu-dyncited by 19

On the map: Diffusion models → Flow matching, time series and science

Abstract (arXiv)

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.

Key facts. Cited by 19 papers in the map (data through 7 Oct 2026). Placed in Flow matching, time series and science, one of 14 directions in Diffusion models. Builds on 83 of its 127 references that are in the map. First posted on arXiv 3 Jul 2025.

Builds on

83 of 127 references.

Closest ancestors · 11

Same direction, newest first

  1. 2025Appa: Bending Weather Dynamics with Latent Diffusion Models for Global Data Assimilation25 Apr
  2. 2025Generative Latent Neural PDE Solver using Flow Matching28 Mar
  3. 2024On conditional diffusion models for PDE simulations21 Oct
  4. 2024Text2PDE: Latent Diffusion Models for Accessible Physics Simulation2 Oct
  5. 2024DiffusionPDE: Generative PDE-Solving Under Partial Observation25 Jun
  6. 2023Benchmarking Autoregressive Conditional Diffusion Models for Turbulent Flow Simulation4 Sep
  7. 2023PDE-Refiner: Achieving Accurate Long Rollouts with Neural PDE Solvers10 Aug
  8. 2023PreDiff: Precipitation Nowcasting with Latent Diffusion Models19 Jul
  9. 2023DYffusion: A Dynamics-informed Diffusion Model for Spatiotemporal Forecasting3 Jun
  10. 2022Flow Matching for Generative Modeling6 Oct
  11. 2022Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow7 Sep

Other references · 72

  1. 2017Attention Is All You Needcited by 59,932
  2. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  3. 2014Adam: A Method for Stochastic Optimizationcited by 36,960
  4. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
  5. 2020Denoising Diffusion Probabilistic Modelscited by 20,851
  6. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,694
  7. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,029
  8. 2020Denoising Diffusion Implicit Modelscited by 9,655
  9. 2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 8,964
  10. 2021Masked Autoencoders Are Scalable Vision Learnerscited by 7,339
  11. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,005
  12. 2022Scalable Diffusion Models with Transformerscited by 5,931
38 more
  1. 2022Generative Adversarial Networkscited by 5,603
  2. 2016Layer Normalizationcited by 5,579
  3. 2021RoFormer: Enhanced Transformer with Rotary Position Embeddingcited by 4,581
  4. 2024Scaling Rectified Flow Transformers for High-Resolution Image Synthesiscited by 4,094
  5. 2019Generative Modeling by Estimating Gradients of the Data Distributioncited by 4,067
  6. 2020Taming Transformers for High-Resolution Image Synthesiscited by 3,229
  7. 2022Elucidating the Design Space of Diffusion-Based Generative Modelscited by 2,994
  8. 2020Fourier Neural Operator for Parametric Partial Differential Equationscited by 2,806
  9. 2022Video Diffusion Modelscited by 2,280
  10. 2022Diffusion Posterior Sampling for General Noisy Inverse Problemscited by 1,375
  11. 2017Sigmoid-Weighted Linear Units for Neural Network Function Approximation in Reinforcement Learningcited by 935
  12. 2020Learning to Simulate Complex Physics with Graph Networkscited by 867
  13. 2020Learning Mesh-Based Simulation with Graph Networkscited by 840
  14. 2018CCNet: Criss-Cross Attention for Semantic Segmentationcited by 717
  15. 2024Movie Gen: A Cast of Media Foundation Modelscited by 614
  16. 2022Fast Sampling of Diffusion Models with Exponential Integratorcited by 526
  17. 2017Deep learning for universal linear embeddings of nonlinear dynamicscited by 491
  18. 2024Revisiting Feature Prediction for Learning Visual Representations from Videocited by 490
  19. 2018Shampoo: Preconditioned Stochastic Tensor Optimizationcited by 413
  20. 2023Analyzing and Improving the Training Dynamics of Diffusion Modelscited by 396
  21. 2025Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Modelscited by 376
  22. 2022Message Passing Neural PDE Solverscited by 368
  23. 2019Axial Attention in Multidimensional Transformerscited by 350
  24. 2022MAGVIT: Masked Generative Video Transformercited by 341
  25. 2023GNOT: A General Neural Operator Transformer for Operator Learningcited by 321
  26. 2023Common Diffusion Noise Schedules and Sample Steps are Flawedcited by 308
  27. 2022MCVD: Masked Conditional Video Diffusion for Prediction, Generation, and Interpolationcited by 307
  28. 2020Query-Key Normalization for Transformerscited by 266
  29. 2022Neighborhood Attention Transformercited by 235
  30. 2024Rotary Position Embedding for Vision Transformercited by 226
  31. 2024Deep Compression Autoencoder for Efficient High-Resolution Diffusion Modelscited by 209
  32. 2023Convolutional Neural Operators for robust and accurate learning of PDEscited by 207
  33. 2019Generalized Sliced Wasserstein Distancescited by 205
  34. 2024Poseidon: Efficient Foundation Models for PDEscited by 186
  35. 2023Fast Training of Diffusion Models with Masked Transformerscited by 172
  36. 2020A Theory of Usable Information Under Computational Constraintscited by 153
  37. 2024The Well: a Large-Scale Collection of Diverse Physics Simulations for Machine Learningcited by 134
  38. 2016Accelerating Eulerian Fluid Simulation With Convolutional Networkscited by 128

50 most cited of 72 All 72 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 5; 2026: 14.

19 papers.

Sort:

By area:

  1. 2025Walrus: A Cross-Domain Foundation Model for Continuum Dynamicscited by 26
  2. 2025Training-Free Data Assimilation with GenCastcited by 5
  3. 2026Probabilistic Retrofitting of Learned Simulatorscited by 5
  4. 2025Predicting partially observable dynamical systems via diffusion models with a multiscale inference schemecited by 4
  5. 2025Reframing Generative Models for Physical Systems using Stochastic Interpolantscited by 2
  6. 2025Efficient Generative Transformer Operators For Million-Point PDEscited by 2
  7. 2026Training-Free Bayesian Filtering with Generative Emulatorscited by 2
  8. 2026TRIE: An Evaluation Framework for Stochastic PDE Surrogatescited by 1
  9. 2026Generative Neural Operators through Diffusion Last Layercited by 0
  10. 2026Inpainting physics: self-supervised learning for context-driven fluid simulationcited by 0
  11. 2026Generating Symmetric Materials using Latent Flow Matchingcited by 0
  12. 2026Wavelet Flow Matching for Multi-Scale Physics Emulationcited by 0
7 more
  1. 2026Reliability of Probabilistic Emulation of Physical Systemscited by 0
  2. 2026Pixel-Space Diffusion Transformerscited by 0
  3. 2026Kastor: An efficient fine-tuning strategy for generative emulation of PDE simulationscited by 0
  4. 2026Geometry-aware Latent Autoregressive Generative Model for PDEs in Complex Domainscited by 0
  5. 2026Beyond Compression: Training Latent Representations for Stable Long-Horizon Rollout in Neural Surrogate Solverscited by 0
  6. 2026CasEm: A Cascade Architecture for Long-Horizon Neural Emulationcited by 0
  7. 2026EnJoi: Ensemble Joint Score Filter for Generative Data Assimilationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Just-in-Time: Training-Free Spatial Acceleration for Diffusion Transformers
  2. 2025LikePhys: Evaluating Intuitive Physics Understanding in Video Diffusion Models via Likelihood Preference
  3. 2026Latent Generative Solvers for Generalizable Long-Term Physics Simulation
  4. 2026Physically Based Rendering in the Latent Space
  5. 2025FunDiff: Diffusion Models over Function Spaces for Physics-Informed Generative Modeling
  6. 2026Inference-time Physics Alignment of Video Generative Models with Latent World Models
  7. 2026PhysLDM: Latent Diffusion for High-Fidelity Deformable Simulation
  8. 2026Diffusing in the Right Space: A Systematic Study of Latent Diffusability