Uncertainty Estimation with Infinitesimal Jackknife, Its Distribution and Mean-Field Approximation

This repository contains a PyTorch implementation of paper "Uncertainty Estimation with Infinitesimal Jackknife, Its Distribution and Mean-Field Approximation".

Code Structure

• Utils/ common util functions for hessian computation, uncertainty metrics evaluation, and etc.
• MNIST/ experiments on MNIST dataset.
• ImageNet/ experiments on ImageNet dataset.

Prerequisites

The following packages are required to run the scripts:

• PyTorch-1.5.0 (or higher) and torchvision
• ipdb
• matplotlib
• seaborn