Robust Machine Comprehension Models via Adversarial Training (N18-2)

Copied to clipboard

Challenge: Existing models for the Stanford Question Answering Dataset suffer from a 50% decrease in F1 score during adversarial evaluation based on AddSent.
Approach: They propose an alternative adversary-generation algorithm, AddSentDiverse, that significantly increases the variance within the adversarial training data by providing effective examples that punish the model for making certain superficial assumptions.
Outcome: The proposed algorithm can achieve a 36.5% increase in F1 score while maintaining performance on the regular SQuAD task.

Similar Papers

Towards Robust Extractive Question Answering Models: Rethinking the Training Methodology (2024.findings-emnlp)

Copied to clipboard

Challenge: Existing models lack robustness against distribution shifts and adversarial attacks when training on unanswerable questions in EQA datasets.
Approach: They propose a novel loss function for the EQA problem to improve the robustness of extractive question answering models by adding adversarial questions to a crowdsourcing process.
Outcome: The proposed method maintains in-domain performance while improving on out-of-domain datasets.
Evaluating and Enhancing the Robustness of Neural Network-based Dependency Parsing Models with Adversarial Examples (2020.acl-main)

Copied to clipboard

Challenge: Previously studies focused on semantic tasks such as sentiment analysis, question answering and reading comprehension.
Approach: They propose two approaches to study where and how adversarial examples exist in dependency parsing . they use a state-of-the-art parser to find adversarials in existing texts .
Outcome: The proposed approaches show that adversarial examples exist in dependency parsing . they show that up to 77% of input examples admit adversarials .
Evaluating Neural Model Robustness for Machine Comprehension (2021.eacl-main)

Copied to clipboard

Challenge: evaluating model robustness to adversarial attacks can provide deeper understanding of how deep neural networks work and what kind of linguistic information is actually captured by neural networks.
Approach: They propose a method for strategic sentence-level perturbations to evaluate model robustness to adversarial attacks using character and word perturbations.
Outcome: The proposed model improves model performance during adversarial attacks by using ensembles and predicts errors in adversarials.
The Impacts of Unanswerable Questions on the Robustness of Machine Reading Comprehension Models (2023.eacl-main)

Copied to clipboard

Challenge: Pretrained language models have achieved super-human performances on many Machine Reading Comprehension (MRC) benchmarks.
Approach: They propose to fine-tune three state-of-the-art language models on SQuAD 1.1 or SQu AD 2.0 and then evaluate their robustness under adversarial attacks.
Outcome: The proposed model is able to perform better under adversarial attacks than model fine-tuned on SQuAD 1.1 or 2.0.
Improving Question Answering Model Robustness with Synthetic Adversarial Data Generation (2021.emnlp-main)

Copied to clipboard

Challenge: a new approach to generate adversarial data is needed to improve question answering models . crowdworkers can fool a model only 8.8% of the time, compared to 17.6% for a trained model without synthetic data.
Approach: They develop a pipeline that generates questions and then filters or labels them to improve quality.
Outcome: The proposed approach improves state-of-the-art on a human-written adversarial dataset by 3.7F1 and improves model generalisation on nine of the twelve MRQA datasets.
Benchmarking Robustness of Machine Reading Comprehension Models (2021.findings-acl)

Copied to clipboard

Challenge: Existing benchmarks only evaluate models' robustness under test-time perturbations or adversarial attacks.
Approach: They propose a model-agnostic benchmark to evaluate models' robustness under adversarial attacks.
Outcome: The proposed model-agnostic benchmark evaluates models under four different types of adversarial attacks.
Undersensitivity in Neural Reading Comprehension (2020.findings-emnlp)

Copied to clipboard

Challenge: Existing models generalise well to in-distribution test sets, yet perform poorly on adversarially selected data.
Approach: They propose an adversarial attack which searches among semantic variations of the question for which a model erroneously predicts the same answer, and with even higher probability.
Outcome: The proposed attack reduces the vulnerability of models trained on SQuAD2.0 and NewsQA, and outperforms a conventional model by as much as 10.9% F1.
Impact of Adversarial Training on Robustness and Generalizability of Language Models (2023.findings-acl)

Copied to clipboard

Challenge: Adversarial training is widely acknowledged as the most effective defense against adversarial attacks, but achieving both robustness and generalization requires a trade-off.
Approach: They propose to compare pre-training data augmentation and training time input perturbations with embedding space perturbations to find out whether they improve generalization.
Outcome: The proposed methods improve generalization and robustness of the trained models.
On the Efficacy of Adversarial Data Collection for Question Answering: Results from a Large-Scale Randomized Study (2021.acl-long)

Copied to clipboard

Challenge: Existing studies have shown that adversarial data collection (ADC) models perform better on other adversarially collected data but are liable under plausible domain shifts.
Approach: They conduct a large-scale controlled study on question answering by assigning workers at random to compose questions either adversarially (with a model in the loop) or in the standard fashion (without a modeling).
Outcome: The proposed model performs better on other adversarial datasets but worse on diverse collection of out-of-domain evaluation sets.
Tougher Text, Smarter Models: Raising the Bar for Adversarial Defence Benchmarks (2025.coling-main)

Copied to clipboard

Challenge: Recent advances in natural language processing have highlighted the vulnerability of deep learning models to adversarial attacks.
Approach: They propose a benchmark for textual adversarial defence that evaluates state-of-the-art defence mechanisms across diverse datasets, models, and tasks.
Outcome: The proposed benchmark incorporates a wide range of datasets and evaluates state-of-the-art defence mechanisms.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations