Not All Samples Are Equal: Quantifying Instance-level Difficulty in Targeted Data Poisoning

Open in new window