When is dataset cartography ineffective? Using training dynamics does not improve robustness against Adversarial SQuAD

Fuente: arXiv
Saved in:
Bibliographic Details
Main Author: Mandal, Paul K.
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915211436359680
author Mandal, Paul K.
author_facet Mandal, Paul K.
contents In this paper, I investigate the effectiveness of dataset cartography for extractive question answering on the SQuAD dataset. I begin by analyzing annotation artifacts in SQuAD and evaluate the impact of two adversarial datasets, AddSent and AddOneSent, on an ELECTRA-small model. Using training dynamics, I partition SQuAD into easy-to-learn, ambiguous, and hard-to-learn subsets. I then compare the performance of models trained on these subsets to those trained on randomly selected samples of equal size. Results show that training on cartography-based subsets does not improve generalization to the SQuAD validation set or the AddSent adversarial set. While the hard-to-learn subset yields a slightly higher F1 score on the AddOneSent dataset, the overall gains are limited. These findings suggest that dataset cartography provides little benefit for adversarial robustness in SQuAD-style QA tasks. I conclude by comparing these results to prior findings on SNLI and discuss possible reasons for the observed differences.
format Preprint
id arxiv_https___arxiv_org_abs_2503_18290
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle When is dataset cartography ineffective? Using training dynamics does not improve robustness against Adversarial SQuAD
Mandal, Paul K.
Computation and Language
Artificial Intelligence
I.2.7; I.2.6; I.5.1
In this paper, I investigate the effectiveness of dataset cartography for extractive question answering on the SQuAD dataset. I begin by analyzing annotation artifacts in SQuAD and evaluate the impact of two adversarial datasets, AddSent and AddOneSent, on an ELECTRA-small model. Using training dynamics, I partition SQuAD into easy-to-learn, ambiguous, and hard-to-learn subsets. I then compare the performance of models trained on these subsets to those trained on randomly selected samples of equal size. Results show that training on cartography-based subsets does not improve generalization to the SQuAD validation set or the AddSent adversarial set. While the hard-to-learn subset yields a slightly higher F1 score on the AddOneSent dataset, the overall gains are limited. These findings suggest that dataset cartography provides little benefit for adversarial robustness in SQuAD-style QA tasks. I conclude by comparing these results to prior findings on SNLI and discuss possible reasons for the observed differences.
title When is dataset cartography ineffective? Using training dynamics does not improve robustness against Adversarial SQuAD
topic Computation and Language
Artificial Intelligence
I.2.7; I.2.6; I.5.1
url https://arxiv.org/abs/2503.18290