Downscaled MNIST

Data for benchmarking machine learning models, taken from Better than classical? The subtle art of benchmarking quantum machine learning models. The Downscaled MNIST classification task is a simplified version of the famous MNIST handwritten digits dataset. This version involves distinguishing between digits 3 and 5 rather than the full range 0-9.

Description of the dataset

This collection provides 19 datasets that consist of flat input vectors of dimension $d=2,\ldots,20$. The inputs were produced by fitting a PCA dimensionality reduction model on the original MNIST training sets, and using the same model to reduce the images from the test set.

Additional details

• The class labels are defined as -1, 1.
• For each dimension, 11552 labeled points are provided for training and 1902 for testing.
• Please see the Source code tab to check how the data was generated.

Example usage

[ds] = qml.data.load("other", name="mnist-pca")

ds.train['4']['inputs'] # points in 4-dimensional space
ds.test['4']['labels'] # labels for the points above

version 0.1 : initial public release