Relationship between a PILD and an FLD is clarified based on overall accuracy.
A PILD is not certainly equivalent to an FLD if the desired outputs are fixed.
A PILD has nothing in common with an FLD when the desired outputs are changeable.
Accuracies of PILDs are improved by optimal thresholds related to sizes and regions.
The iterative learning strategy of PILDs is proposed, realized and verified.