This dataset was collected in ELL409 (machine learning) taught at IIT Delhi by Prof. Prathosh AP. It is recommended that you train on the original MNIST dataset and test on this provided dataset. This is a good dataset for studying bias correction in deep learning models. The file has 7948 rows with 7947 test samples. The first row is the header row. There are 785 columns. The first 784 are MNIST images. The last column is its label.