ULE is a digit recognition dataset (toy data)

The task of ULE is handwritten digit recognition. The dataset was constructed from the MNIST data that is made available by Yann LeCun and Corinna Cortes. The digits have been size-normalized and centered in a fixed-size image of dimension 28x28. We provide the raw data and the labels to illustrate all the aspects of the Unsupervised and Transfer Learning challenge. THIS DATASET IS NOT PART OF THE EVALUATION DATASETS.
Part of the data of the MNIST dataset was used previously in the Feature Selection challenge under the name GISETTE, and in the Performance Prediction challenge, the Model Selection game, under the name of GINA.
CausalityThis dataset is used in the Unsupervised and Transfer Learning Challenge by the Causality Workbench