Consider a classifier trained on the 60k MNIST images (no augmentation). We test it on images...

 

Transcribed Image Text:

Consider a classifier trained on the 60k MNIST images (no augmentation). We test it on images where the pixels have been shifted. We expect the test accuracy to be poor. Is this an example of distribution shift? Please provide a reason. Consider a classifier trained on the 60k MNIST images (no augmentation). We test it on images where the pixels have been shifted. We expect the test accuracy to be poor. Is this an example of distribution shift? Please provide a reason.

Answer rating: 100% (QA)

It has been given that the classfier is trained in MNIST images dataset So let us first see in short ... View the full answer