Effects of Human Adversarial and Affable Samples on BERT Generalizability
Aparna Elangovan, Estrid He, Yuan Li, Karin M. Verspoor · 2023
BERT-based models have had strong performance on leaderboards, yet have been demonstrably worse in real-world settings requiring generalization.Limited quantities of training data is considered a key impediment to achieving generalizability in machine learning.In this paper, we examine the impact of training data quality, not quantity, on a model's generalizability.We consider two characteristics of training data: the portion of human-adversarial (hadversarial), i.e. sample pairs with seemingly minor differences but different ground-truth labels, and human-affable (h-affable) training samples, i.e. sample pairs with minor differences but the same ground-truth label.We find that for a fixed size of training samples, as a rule of thumb, having 10-30% h-adversarial instances improves the precision, and therefore F 1 , by up to 20 points in the tasks of text classification and relation extraction.Increasing h-adversarials beyond this range can result in performance plateaus or even degradation.In contrast, h-affables may not contribute to a model's generalizability and may even degrade generalization performance.