Analyzing Inverse Problems with Invertible Neural Networks

Abstract

In many tasks, in particular in natural science, the goal is to determine hidden system parameters from a set of measurements. Often, the forward process from parameter- to measurement-space is a well-defined function, whereas the inverse problem is ambiguous: one measurement may map to multiple different sets of parameters. In this setting, the posterior parameter distribution, conditioned on an input measurement, has to be determined. We argue that a particular class of neural networks is well suited for this task -- so-called Invertible Neural Networks (INNs). Although INNs are not new, they have, so far, received little attention in literature. While classical neural networks attempt to solve the ambiguous inverse problem directly, INNs are able to learn it jointly with the well-defined forward process, using additional latent output variables to capture the information otherwise lost. Given a specific measurement and sampled latent variables, the inverse pass of the INN provides a full distribution over parameter space. We verify experimentally, on artificial data and real-world problems from astrophysics and medicine, that INNs are a powerful analysis tool to find multi-modalities in parameter space, to uncover parameter correlations, and to identify unrecoverable parameters.

Cited by in corpus (58)

• Guided Image Generation with Conditional Invertible Neural Networks
• Invertible Residual Networks
• Neural SDE: Stabilizing Neural ODE Networks with Stochastic Noise
• Solving inverse problems using conditional invertible neural networks
• Conditional Flow Variational Autoencoders for Structured Sequence Prediction
• Excessive Invariance Causes Adversarial Vulnerability
• Invertible generative models for inverse problems: mitigating representation error and dataset bias
• Coupling-based Invertible Neural Networks Are Universal Diffeomorphism Approximators
• Machine learning for complete intersection Calabi-Yau manifolds: a methodological study
• Stabilizing Invertible Neural Networks Using Mixture Models
• Uncertainty-aware performance assessment of optical imaging modalities with invertible neural networks
• Stochastic Normalizing Flows for Inverse Problems: a Markov Chains Viewpoint
• Parrot: Data-Driven Behavioral Priors for Reinforcement Learning
• Benchmarking Invertible Architectures on Inverse Problems
• DeepFlow: History Matching in the Space of Deep Generative Models
• A Review of Single-Source Deep Unsupervised Visual Domain Adaptation
• Benchmarking deep inverse models over time, and the neural-adjoint method
• Modeling Lost Information in Lossy Image Compression
• Some models are useful, but how do we know which ones? Towards a unified Bayesian model taxonomy
• Understanding and Mitigating Exploding Inverses in Invertible Neural Networks
• A unified representation network for segmentation with missing modalities
• Deep learning calibration of option pricing models: some pitfalls and solutions
• Neural Empirical Bayes: Source Distribution Estimation and its Applications to Simulation-Based Inference
• Large-Scale Wasserstein Gradient Flows
• A Flow-based Truncated Denoising Diffusion Model for Super-resolution Magnetic Resonance Spectroscopic Imaging
• Enhanced Invertible Encoding for Learned Image Compression
• InfoCNF: An Efficient Conditional Continuous Normalizing Flow with Adaptive Solvers
• Deep Involutive Generative Models for Neural MCMC
• Deconvolutional Networks on Graph Data
• Towards Interpretable Polyphonic Transcription with Invertible Neural Networks
• Latent Normalizing Flows for Many-to-Many Cross-Domain Mappings
• Bayesian Active Learning for Structured Output Design
• A Method to Model Conditional Distributions with Normalizing Flows
• InFlow: Robust outlier detection utilizing Normalizing Flows
• Comparison of Distal Teacher Learning with Numerical and Analytical Methods to Solve Inverse Kinematics for Rigid-Body Mechanisms
• Generative Design of a Gas Turbine Combustor Using Invertible Neural Networks
• Parallelised Diffeomorphic Sampling-based Motion Planning
• Representing Ambiguity in Registration Problems with Conditional Invertible Neural Networks
• Invertible Zero-Shot Recognition Flows
• Invertible Image Rescaling
• Mixture Manifold Networks: A Computationally Efficient Baseline for Inverse Modeling
• Learning Bloch Simulations for MR Fingerprinting by Invertible Neural Networks
• A Survey on Machine Learning Applied to Dynamic Physical Systems
• Continuous Generative Neural Networks: A Wavelet-Based Architecture in Function Spaces
• Neural Group Actions
• Invertible Neural BRDF for Object Inverse Rendering
• Digital phase-only holography using deep conditional generative models
• Improved Surrogates in Inertial Confinement Fusion with Manifold and Cycle Consistencies
• Learning ODEs via Diffeomorphisms for Fast and Robust Integration
• A Temporal Kernel Approach for Deep Learning with Continuous-time Information
• On the Generative Utility of Cyclic Conditionals
• Bundle Networks: Fiber Bundles, Local Trivializations, and a Generative Approach to Exploring Many-to-one Maps
• Solving Inverse Problems for Spectral Energy Distributions with Deep Generative Networks
• Mitigating Generation Shifts for Generalized Zero-Shot Learning
• Learning to Shift Attention for Motion Generation
• Uncertainty quantification for ptychography using normalizing flows
• Conditional Invertible Neural Networks for Diverse Image-to-Image Translation
• Invertible Surrogate Models: Joint surrogate modelling and reconstruction of Laser-Wakefield Acceleration by invertible neural networks