Tile/copy our data to increase the training set size and reduce the number of training epochs.

How does duplicating data help?

It does not improve the performance of the trained models in this course.

However, copying data N times can mean we train the model N times within one epoch, instead of N epochs. The benefit of training less epochs includes reducing the overheads in switching from one epoch to the next, such as computing the validation scores. In other words, we spend less time.


How is the data set size increased all of a sudden? Is it just repeating?