Image-to-Image Regression with Distribution-Free Uncertainty Quantification and Applications in Imaging (2022-02-10T00:00:00.000000Z)

TL;DR

Uncertainty quantification techniques are developed with rigorous statistical guarantees for image-to-image regression problems that derive uncertainty intervals around each pixel that are guaranteed to contain the true value with a user-specified confidence probability.

Abstract

Image-to-image regression is an important learning task, used frequently in biological imaging. Current algorithms, however, do not generally offer statistical guarantees that protect against a model's mistakes and hallucinations. To address this, we develop uncertainty quantification techniques with rigorous statistical guarantees for image-to-image regression problems. In particular, we show how to derive uncertainty intervals around each pixel that are guaranteed to contain the true value with a user-specified confidence probability. Our methods work in conjunction with any base machine learning model, such as a neural network, and endow it with formal mathematical guarantees -- regardless of the true unknown data distribution or choice of model. Furthermore, they are simple to implement and computationally inexpensive. We evaluate our procedure on three image-to-image regression tasks: quantitative phase microscopy, accelerated magnetic resonance imaging, and super-resolution transmission electron microscopy of a Drosophila melanogaster brain.

Authors

Michael I. Jordan

14 papers

T. Alshaabi

2 papers

Anastasios Nikolas Angelopoulos

4 papers

TL;DR

Abstract

Authors

References82 items

Learn then Test: Calibrating Predictive Algorithms to Achieve Risk Control

Fair Conformal Predictors for Applications in Medical Imaging

A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantification

Improving Conditional Coverage via Orthogonal Quantile Regression

Private Prediction Sets

Distribution-Free, Risk-Controlling Prediction Sets

Uncertainty Sets for Image Classifiers using Conformal Prediction

Classification with Valid and Adaptive Coverage

Knowing what You Know: valid and validated confidence sets in multiclass and multilabel prediction

Interval Neural Networks: Uncertainty Scores

Image denoising review: From classical to state-of-the-art approaches

Deep Learning on Image Denoising: An overview

Deep Ensembles: A Loss Landscape Perspective

Quantifying Point-Prediction Uncertainty in Neural Networks via Residual Estimation with an I/O Kernel

Conformalized Quantile Regression

fastMRI: An Open Dataset and Benchmarks for Accelerated MRI

On the Spectral Bias of Neural Networks

Quantitative Phase Imaging and Artificial Intelligence: A Review

Cautious Deep Learning

Attention U-Net: Learning Where to Look for the Pancreas

Uncertainty Estimates and Multi-hypotheses Networks for Optical Flow

Content-aware image restoration: pushing the limits of fluorescence microscopy

A Complete Electron Microscopy Volume of the Brain of Adult Drosophila melanogaster

Neural Style Transfer: A Review

Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks

Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles

Image-to-Image Translation with Conditional Adversarial Networks

Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network

Least Ambiguous Set-Valued Classifiers With Bounded Error Levels

Image Style Transfer Using Convolutional Neural Networks

Colorful Image Colorization

Perceptual Losses for Real-Time Style Transfer and Super-Resolution

Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning

U-Net: Convolutional Networks for Biomedical Image Segmentation

Quantitative differential phase contrast imaging in an LED array microscope.

Image Super-Resolution Using Deep Convolutional Networks

Classification with confidence

Learning a Deep Convolutional Network for Image Super-Resolution

Gradient boosting machines, a tutorial

Distribution-Free Prediction Sets

A conformal prediction approach to explore functional data

Quantitative Phase Imaging

Image denoising: Can plain neural networks compete with BM3D?

Quantitative phase imaging in biomedicine

Estimating conditional quantiles with the help of the pinball loss

Additive models for quantile regression: Model selection and confidence bandaids

Image Super-Resolution Via Sparse Representation

Neural network with ensembles

The Quality of Accounting Information in Politically Connected Firms

Network effects and the dynamics of migration and inequality: Theory and evidence from Mexico

Quantile Regression Forests

Nonparametric Quantile Estimation

Quantile Regression

Elements of Information Theory

A Simple Quantile Regression via Support Vector Machine

A non-local algorithm for image denoising

Counterfactual decomposition of changes in wage distributions using quantile regression

Algorithmic Learning in a Random World

Super-resolution through neighbor embedding

Example-Based Super-Resolution

Machine-Learning Applications of Algorithmic Randomness

Direct use of regression quantiles to construct confidence sets in linear models

Global nonparametric estimation of conditional quantile functions and their derivatives

Cubic convolution interpolation for digital image processing

Approximation properties of the spline fit

Two inequalities implied by unique decipherability

ON POLYA FREQUENCY FUNCTIONS. III. THE POSITIVITY OF TRANSLATION DETERMINANTS WITH AN APPLICATION TO THE INTERPOLATION PROBLEM BY SPLINE CURVES(

Berkeley single cell computational microscopy dataset, 2021

GENERATIVE ADVERSARIAL NETS

U-Net++: A nested U-Net architecture for medical image segmentation," in Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support

Quantile Regression : Penalized from : Handbook of Quantile Regression

Uncertainty in Deep Learning

Regression Quantiles

Statistical Methods in Medical Research

Statistical inference on heteroscedastic models based on regression quantiles

Estimating the mean and variance of the target probability distribution