Score-based Diffusion Models in Function Space

Diffusion models have recently emerged as a powerful framework for generative modeling. They consist of a forward process that perturbs input data with Gaussian white noise and a reverse process that learns a score function to generate samples by denoising. Despite their tremendous success, they are mostly formulated on finite-dimensional spaces, e.g. Euclidean, limiting their applications to many domains where the data has a functional form such as in scientific computing and 3D geometric data analysis. In this work, we introduce a mathematically rigorous framework called Denoising Diffusion Operators (DDOs) for training diffusion models in function space. In DDOs, the forward process perturbs input functions gradually using a Gaussian process. The generative process is formulated by integrating a function-valued Langevin dynamic. Our approach requires an appropriate notion of the score for the perturbed data distribution, which we obtain by generalizing denoising score matching to function spaces that can be infinite-dimensional. We show that the corresponding discretized algorithm generates accurate samples at a fixed cost that is independent of the data resolution. We theoretically and numerically verify the applicability of our approach on a set of problems, including generating solutions to the Navier-Stokes equation viewed as the push-forward distribution of forcings from a Gaussian Random Field (GRF).

[1]  Padhraic Smyth,et al.  Diffusion Generative Models in Infinite Dimensions , 2022, AISTATS.

[2]  Felix Heide,et al.  DiffusionSDF: Conditional Generative Modeling of Signed Distance Functions , 2022, ArXiv.

[3]  Adam M. Oberman,et al.  Score-based Denoising Diffusion with Non-Isotropic Gaussian Noise Models , 2022, ArXiv.

[4]  Rianne van den Berg,et al.  Protein structure generation via folding diffusion , 2022, Nature communications.

[5]  Ben Poole,et al.  DreamFusion: Text-to-3D using 2D Diffusion , 2022, ICLR.

[6]  Valentin De Bortoli,et al.  Spectral Diffusion Processes , 2022, ArXiv.

[7]  Chi-Wing Fu,et al.  Neural Wavelet-domain Diffusion for 3D Shape Generation , 2022, SIGGRAPH Asia.

[8]  E. Hoogeboom,et al.  Blurring Diffusion Models , 2022, ICLR.

[9]  Valentin De Bortoli,et al.  Wavelet Score-Based Generative Modeling , 2022, NeurIPS.

[10]  Walter A. Talbott,et al.  GAUDI: A Neural Architect for Immersive 3D Scene Generation , 2022, NeurIPS.

[11]  A. Solin,et al.  Generative Modelling With Inverse Heat Dissipation , 2022, ICLR.

[12]  Alan D. Saul,et al.  Neural Diffusion Processes , 2022, ArXiv.

[13]  Xiang Lisa Li,et al.  Diffusion-LM Improves Controllable Text Generation , 2022, NeurIPS.

[14]  David J. Fleet,et al.  Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding , 2022, NeurIPS.

[15]  Vikram S. Voleti,et al.  MCVD: Masked Conditional Video Diffusion for Prediction, Generation, and Interpolation , 2022, ArXiv.

[16]  Anima Anandkumar,et al.  Diffusion Models for Adversarial Purification , 2022, ICML.

[17]  K. Azizzadenesheli,et al.  Generative Adversarial Neural Operators , 2022, Trans. Mach. Learn. Res..

[18]  Bowen Jing,et al.  Subspace Diffusion Generative Models , 2022, ECCV.

[19]  A. Stuart,et al.  The Cost-Accuracy Trade-Off In Operator Learning With Neural Networks , 2022, Journal of Machine Learning.

[20]  S. Ermon,et al.  GeoDiff: a Geometric Diffusion Model for Molecular Conformation Generation , 2022, ICLR.

[21]  K. Azizzadenesheli,et al.  FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operators , 2022, ArXiv.

[22]  Danilo Jimenez Rezende,et al.  From data to functa: Your data point is a function and you can treat it like one , 2022, ICML.

[23]  Il-Chul Moon,et al.  Soft Truncation: A Universal Training Technique of Score-based Diffusion Model for High Precision Score Estimation , 2021, ICML.

[24]  Y. Teh,et al.  Generative Models as Distributions of Functions , 2021, AISTATS.

[25]  Nikola B. Kovachki,et al.  Neural Operator: Learning Maps Between Function Spaces , 2021, ArXiv.

[26]  Kamyar Azizzadenesheli,et al.  Seismic wave propagation and inversion with Neural Operators , 2021, The Seismic Record.

[27]  Siddhartha Mishra,et al.  On universal approximation and error bounds for Fourier Neural Operators , 2021, J. Mach. Learn. Res..

[28]  Nikola B. Kovachki,et al.  Learning Dissipative Dynamics in Chaotic Systems , 2021, 2106.06898.

[29]  Jan Kautz,et al.  Score-based Generative Modeling in Latent Space , 2021, NeurIPS.

[30]  Jiajun Wu,et al.  3D Shape Generation and Completion through Point-Voxel Diffusion , 2021, 2021 IEEE/CVF International Conference on Computer Vision (ICCV).

[31]  Richard E. Turner,et al.  The Gaussian Neural Process , 2021, ArXiv.

[32]  Xiaolong Wang,et al.  Learning Continuous Image Representation with Local Implicit Image Function , 2020, 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).

[33]  Victor Lempitsky,et al.  Image Generators with Conditionally-Independent Pixel Synthesis , 2020, 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).

[34]  Abhishek Kumar,et al.  Score-Based Generative Modeling through Stochastic Differential Equations , 2020, ICLR.

[35]  Mohamed Elhoseiny,et al.  Adversarial Generation of Continuous Images , 2020, 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).

[36]  Nikola B. Kovachki,et al.  Fourier Neural Operator for Parametric Partial Differential Equations , 2020, ICLR.

[37]  Jiaming Song,et al.  Denoising Diffusion Implicit Models , 2020, ICLR.

[38]  Bryan Catanzaro,et al.  DiffWave: A Versatile Diffusion Model for Audio Synthesis , 2020, ICLR.

[39]  Nicholas H. Nelsen,et al.  The Random Feature Model for Input-Output Maps between Banach Spaces , 2020, SIAM J. Sci. Comput..

[40]  Pieter Abbeel,et al.  Denoising Diffusion Probabilistic Models , 2020, NeurIPS.

[41]  Pratul P. Srinivasan,et al.  NeRF , 2020, ECCV.

[42]  Kamyar Azizzadenesheli,et al.  Neural Operator: Graph Kernel Network for Partial Differential Equations , 2020, ICLR 2020.

[43]  George Em Karniadakis,et al.  DeepONet: Learning nonlinear operators for identifying differential equations based on the universal approximation theorem of operators , 2019, ArXiv.

[44]  Yang Song,et al.  Generative Modeling by Estimating Gradients of the Data Distribution , 2019, NeurIPS.

[45]  Yee Whye Teh,et al.  Attentive Neural Processes , 2019, ICLR.

[46]  Richard A. Newcombe,et al.  DeepSDF: Learning Continuous Signed Distance Functions for Shape Representation , 2019, 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).

[47]  A. Stuart,et al.  The Bayesian Approach to Inverse Problems , 2013, 1302.6989.

[48]  A. Kupiainen Quantum Fields and Probability , 2016, 1611.05240.

[49]  Surya Ganguli,et al.  Deep Unsupervised Learning using Nonequilibrium Thermodynamics , 2015, ICML.

[50]  Catherine E. Powell,et al.  An Introduction to Computational Stochastic PDEs , 2014 .

[51]  Gary J. Chandler,et al.  Invariant recurrent solutions embedded in a turbulent two-dimensional Kolmogorov flow , 2013, Journal of Fluid Mechanics.

[52]  G. Roberts,et al.  MCMC Methods for Functions: ModifyingOld Algorithms to Make Them Faster , 2012, 1202.0709.

[53]  H. Rue,et al.  An explicit link between Gaussian fields and Gaussian Markov random fields: the stochastic partial differential equation approach , 2011 .

[54]  Pascal Vincent,et al.  A Connection Between Score Matching and Denoising Autoencoders , 2011, Neural Computation.

[55]  M. Pelagatti Stationary Processes , 2011 .

[56]  Andrew M. Stuart,et al.  Inverse problems: A Bayesian perspective , 2010, Acta Numerica.

[57]  Carl E. Rasmussen,et al.  Gaussian processes for machine learning , 2005, Adaptive computation and machine learning.

[58]  R. Shterenberg,et al.  Blow up and regularity for fractal Burgers equation , 2008, 0804.3549.

[59]  G. Prato An Introduction to Infinite-Dimensional Analysis , 2006 .

[60]  Aapo Hyvärinen,et al.  Estimation of Non-Normalized Statistical Models by Score Matching , 2005, J. Mach. Learn. Res..

[61]  M. Gelbrich On a Formula for the L2 Wasserstein Metric between Measures on Euclidean and Hilbert Spaces , 1990 .

[62]  R. Temam Infinite Dimensional Dynamical Systems in Mechanics and Physics Springer Verlag , 1993 .

[63]  H. Föllmer,et al.  Time reversal of infinite-dimensional diffusions , 1986 .

[64]  R. Adler,et al.  The Geometry of Random Fields , 1982 .

[65]  P. Whittle ON STATIONARY PROCESSES IN THE PLANE , 1954 .

[66]  G. M.,et al.  Partial Differential Equations I , 2023, Applied Mathematical Sciences.