Cold Case: The Lost MNIST Digits

TL;DR

This paper reconstructs the original MNIST dataset from the NIST source, providing a more accurate version and a complete test set, enabling analysis of long-term effects of MNIST experiments on classifier performance.

Contribution

It offers a detailed reconstruction of MNIST's derivation process and a complete test set, facilitating more accurate benchmarking and historical analysis.

Findings

Reconstructed MNIST closely matches original accuracy levels.

Long-term MNIST experiments have minimal impact on classifier rankings.

Classifier comparisons remain reliable despite slight accuracy differences.

Abstract

Although the popular MNIST dataset [LeCun et al., 1994] is derived from the NIST database [Grother and Hanaoka, 1995], the precise processing steps for this derivation have been lost to time. We propose a reconstruction that is accurate enough to serve as a replacement for the MNIST dataset, with insignificant changes in accuracy. We trace each MNIST digit to its NIST source and its rich metadata such as writer identifier, partition identifier, etc. We also reconstruct the complete MNIST test set with 60,000 samples instead of the usual 10,000. Since the balance 50,000 were never distributed, they enable us to investigate the impact of twenty-five years of MNIST experiments on the reported testing performances. Our results unambiguously confirm the trends observed by Recht et al. [2018, 2019]: although the misclassification rates are slightly off, classifier ordering and model selection remain broadly reliable. We attribute this phenomenon to the pairing benefits of comparing classifiers on the same digits.