ML researchers will sometimes vary the size of the training data set to see what happens. It’s not common - except in scaling law research. But it’s never called “ablation”.