Robust Training under Linguistic Adversity

Li, Yitong; Cohn, Trevor; Baldwin, Timothy

doi:10.18653/v1/e17-2004

Cited by 52 publications

(38 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The notion of linguistically plausible corruption is also explored by Li et al (2017), who created adversarial examples with syntactic and semantic noise (reordering and word substitutions respectively). When training with these noisy datasets, they obtained better performance on several text classification tasks.…”

Section: Related Workmentioning

confidence: 99%

Neural Machine Translation of Text from Non-Native Speakers

Anastasopoulos

Lui²,

Nguyen

et al. 2019

Proceedings of the 2019 Conference of the North

View full text Add to dashboard Cite

Neural Machine Translation (NMT) systems are known to degrade when confronted with noisy data, especially when the system is trained only on clean data. In this paper, we show that augmenting training data with sentences containing artificially-introduced grammatical errors can make the system more robust to such errors. In combination with an automatic grammar error correction system, we can recover 1.9 BLEU out of 3.1 BLEU lost due to grammatical errors. We also present a set of Spanish translations of the JFLEG grammar error correction corpus, which allows for testing NMT robustness to real grammatical errors.

show abstract

Section: Related Workmentioning

confidence: 99%

Neural Machine Translation of Text from Non-Native Speakers

Anastasopoulos

Lui²,

Nguyen

et al. 2019

Proceedings of the 2019 Conference of the North

View full text Add to dashboard Cite

show abstract

“…These adversarial attacks highlight often unintuitive model failure modes and present a challenge to deploying NLP models. Common attempts to mitigate the issue are adversarial training (Ebrahimi et al, 2018) and data augmentation (Belinkov and Bisk, 2018;Li et al, 2017), which lead to improved accuracy on adversarial examples. However, this might cause a false sense of security, as there is generally no guarantee that stronger adversaries could not circumvent defenses to find other successful attacks (Carlini and Wagner, 2017;Athalye et al, 2018;.…”

Section: Introductionmentioning

confidence: 99%

Achieving Verified Robustness to Symbol Substitutions via Interval Bound Propagation

Huang¹,

Stanforth²,

Welbl³

et al. 2019

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conferen

100

116

View full text Add to dashboard Cite

Neural networks are part of many contemporary NLP systems, yet their empirical successes come at the price of vulnerability to adversarial attacks. Previous work has used adversarial training and data augmentation to partially mitigate such brittleness, but these are unlikely to find worst-case adversaries due to the complexity of the search space arising from discrete text perturbations. In this work, we approach the problem from the opposite direction: to formally verify a system's robustness against a predefined class of adversarial attacks. We study text classification under synonym replacements or character flip perturbations. We propose modeling these input perturbations as a simplex and then using Interval Bound Propagation -a formal model verification method. We modify the conventional loglikelihood training objective to train models that can be efficiently verified, which would otherwise come with exponential search complexity. The resulting models show only little difference in terms of nominal accuracy, but have much improved verified accuracy under perturbations and come with an efficiently computable formal guarantee on worst case adversaries.

show abstract

“…The first approach draws on our earlier work on learning robust text representations (Li et al, 2017), and is based on synonyms of the given token, based on Princeton WordNet (Miller et al, 1990) using the NLTK API (Bird, 2006). Here, we test possible synonyms, considering their part-ofspeech tag, asking the system s whether the loss is reduced after substitution.…”

Section: Token Substitution Methodsmentioning

confidence: 99%

BIBI System Description: Building with CNNs and Breaking with Deep Reinforcement Learning

Li¹,

Cohn²,

Baldwin³

2017

Proceedings of the First Workshop on Building Linguistically Generalizable NLP Systems

Self Cite

View full text Add to dashboard Cite

This paper describes our submission to the sentiment analysis sub-task of "Build It, Break It: The Language Edition (BIBI)", on both the builder and breaker sides. As a builder, we use convolutional neural nets, trained on both phrase and sentence data. As a breaker, we use Q-learning to learn minimal change pairs, and apply a token substitution method automatically. We analyse the results to gauge the robustness of NLP systems.

show abstract

Robust Training under Linguistic Adversity

Cited by 52 publications

References 27 publications

Neural Machine Translation of Text from Non-Native Speakers

Neural Machine Translation of Text from Non-Native Speakers

Achieving Verified Robustness to Symbol Substitutions via Interval Bound Propagation

BIBI System Description: Building with CNNs and Breaking with Deep Reinforcement Learning

Contact Info

Product

Resources

About