Abstract:We propose a novel modular inference approach combining two different generative models — generative adversarial networks (GAN) and normalizing flows — to approximate the posterior distribution of physics-based Bayesian inverse problems framed in high-dimensional ambient spaces. We dub the proposed framework GAN-Flow. The proposed method leverages the intrinsic dimension reduction and superior sample generation capabilities of GANs to define a low-dimensional data-driven prior distribution. Once a trained GAN-prior is available, the inverse problem is solved entirely in the latent space of the GAN using variational Bayesian inference with normalizing flow-based variational distribution, which approximates low-dimensional posterior distribution by transforming realizations from the low-dimensional latent prior (Gaussian) to corresponding realizations of a low-dimensional variational posterior distribution. The trained GAN generator then maps realizations from this approximate posterior distribution in the latent space back to the high-dimensional ambient space. We also propose a two-stage training strategy for GAN-Flow wherein we train the two generative models sequentially. Thereafter, GAN-Flow can estimate the statistics of posterior-predictive quantities of interest at virtually no additional computational cost. The synergy between the two types of generative models allows us to overcome many challenges associated with the application of Bayesian inference to large-scale inverse problems, chief among which are describing an informative prior and sampling from the high-dimensional posterior. GAN-Flow does not involve Markov chain Monte Carlo simulation, making it particularly suitable for solving large-scale inverse problems. We demonstrate the efficacy and flexibility of GAN-Flow on various physics-based inverse problems of varying ambient dimensionality and prior knowledge using different types of GANs and normalizing flows. Notably, one of the applications we consider involves a 65,536-dimensional inverse problem of phase retrieval wherein an object is reconstructed from sparse noisy measurements of the magnitude of its Fourier transform.

PCSAGAN: a physics-constrained generative network based on self-attention for high-fidelity flow field reconstruction

3D-Aware Image Synthesis Via Learning Structural and Textural Representations

A generative adversarial network based on an efficient transformer for high-fidelity flow field reconstruction

Flow Reconstruction with Uncertainty Quantification from Noisy Measurements Based on Bayesian Physics-Informed Neural Networks

Flow Reconstruction Using Spatially Restricted Domains Based on Enhanced Super-Resolution Generative Adversarial Networks

Generate Optical Flow with Conditional Generative Adversarial Network

Sag-flownet: self-attention generative network for airfoil flow field prediction

Flow reconstruction in time-varying geometries using graph neural networks

Navier–stokes Generative Adversarial Network: a physics-informed deep learning model for fluid flow generation

An Adversarial Non-Volume Preserving Flow Model with Boltzmann Priors

A dimension-reduced variational approach for solving physics-based inverse problems using generative adversarial network priors and normalizing flows

Self-Assessed Generation: Trustworthy Label Generation for Optical Flow and Stereo Matching in Real-world

Using Physics-Informed Generative Adversarial Networks to Perform Super-Resolution for Multiphase Fluid Simulations

Boundary-aware Decoupled Flow Networks for Realistic Extreme Rescaling

Multi-scale data reconstruction of turbulent rotating flows with Gappy POD, Extended POD and Generative Adversarial Networks

Physics-informed Recurrent Super-Resolution Generative Reconstruction in Rotating Detonation Combustor

A 3D super-resolution of wind fields via physics-informed pixel-wise self-attention generative adversarial network

Physics-Informed Generative Adversarial Network Solution to Buckley–Leverett Equation

Data and physics-driven modeling for fluid flow with a physics-informed graph convolutional neural network

Stochastic reconstruction of porous media based on attention mechanisms and multi-stage generative adversarial network

TransFlowNet: A Physics-Constrained Transformer Framework for Spatio-Temporal Super-Resolution of Flow Simulations.