Is iterative pruning always structured?

Iterative pruning works like the following:

1. Evaluate model performance, continue if performance is fine
2. Apply pruning
3. Train the model
4. Go back to step 1

When doing this process using unstructured pruning I set 5% of the weights, which absolute value is closest to zero, to zero.

After that I train the model. The problem is that after the training the 5% of the weights which were set to zero are again very close to zero. So after applying pruning the second time the same weights are set to zero as the first time.

Is there such a thing as iterative unstructured pruning and if yes, how should I do it?