Normal vs. Adversarial: Salience-based Analysis of Adversarial Samples for Relation Extraction

Li, Luoqiu; Chen, Xiang; Bi, Zhongnan; Xie, Xing; Deng, Shumin; Zhang, Ningyu; Tan, Chuanqi; Chen, Mosha; Chen, Huajun

doi:10.48550/arxiv.2104.00312

Cited by 2 publications

(2 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In some scenarios, gradient-based approaches were shown to provide more faithful explanations than attention-based methods [15]. This family of gradient-based explainability methods have been applied [16,17,30], yet in a task-specific manner, to different downstream tasks.…”

Section: Related Workmentioning

confidence: 99%

Grad-SAM: Explaining Transformers via Gradient Self-Attention Maps

Barkan,

Hauon,

Caciularu

et al. 2022

Preprint

View full text Add to dashboard Cite

Transformer-based language models significantly advanced the stateof-the-art in many linguistic tasks. As this revolution continues, the ability to explain model predictions has become a major area of interest for the NLP community. In this work, we present Gradient Self-Attention Maps (Grad-SAM) -a novel gradient-based method that analyzes self-attention units and identifies the input elements that explain the model's prediction the best. Extensive evaluations on various benchmarks show that Grad-SAM obtains significant improvements over state-of-the-art alternatives.

show abstract

Section: Related Workmentioning

confidence: 99%

Grad-SAM: Explaining Transformers via Gradient Self-Attention Maps

Barkan,

Hauon,

Caciularu

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…However, there is a significant gap between task performance and model generalizability. Previous approaches have indicated that neural models suffer from poor robustness when encountering randomly permuted contexts [22] and adversarial examples [11,13].…”

Section: Introductionmentioning

confidence: 99%

Disentangled Contrastive Learning for Learning Robust Textual Representations

Chen¹,

Xie²,

Bi³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Although the self-supervised pre-training of transformer models has resulted in the revolutionizing of natural language processing (NLP) applications and the achievement of state-of-the-art results with regard to various benchmarks, this process is still vulnerable to small and imperceptible permutations originating from legitimate inputs. Intuitively, the representations should be similar in the feature space with subtle input permutations, while large variations occur with different meanings. This motivates us to investigate the learning of robust textual representation in a contrastive manner. However, it is non-trivial to obtain opposing semantic instances for textual samples. In this study, we propose a disentangled contrastive learning method that separately optimizes the uniformity and alignment of representations without negative sampling. Specifically, we introduce the concept of momentum representation consistency to align features and leverage power normalization while conforming the uniformity. Our experimental results for the NLP benchmarks demonstrate that our approach can obtain better results compared with the baselines, as well as achieve promising improvements with invariance tests and adversarial attacks. The code is available in https://github.com/zjunlp/DCL.

show abstract

Normal vs. Adversarial: Salience-based Analysis of Adversarial Samples for Relation Extraction

Cited by 2 publications

References 0 publications

Grad-SAM: Explaining Transformers via Gradient Self-Attention Maps

Grad-SAM: Explaining Transformers via Gradient Self-Attention Maps

Disentangled Contrastive Learning for Learning Robust Textual Representations

Contact Info

Product

Resources

About