{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/adversarial-self-attention-for-language","title":"Adversarial Self-Attention for Language Understanding","arxiv_id":"2206.12608","date":"2022-06-25","proceeding":null,"authors":["Hongqiu Wu","Ruixue Ding","Hai Zhao","Pengjun Xie","Fei Huang","Min Zhang"],"abstract":"Deep neural models (e.g. Transformer) naturally learn spurious features, which create a ``shortcut'' between the labels and inputs, thus impairing the generalization and robustness. This paper advances the self-attention mechanism to its robust variant for Transformer-based pre-trained language models (e.g. BERT). We propose \\textit{Adversarial Self-Attention} mechanism (ASA), which adversarially biases the attentions to effectively suppress the model reliance on features (e.g. specific keywords) and encourage its exploration of broader semantics. We conduct a comprehensive evaluation across a wide range of tasks for both pre-training and fine-tuning stages. For pre-training, ASA unfolds remarkable performance gains compared to naive training for longer steps. For fine-tuning, ASA-empowered models outweigh naive models by a large margin considering both generalization and robustness.","url_abs":"https://arxiv.org/abs/2206.12608v3","url_pdf":"https://arxiv.org/pdf/2206.12608v3.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"adversarial-self-attention-for-language","repo_url":"https://github.com/gingasan/adversarialsa","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"machine-reading-comprehension","task_name":"Machine Reading Comprehension"},{"task_slug":"named-entity-recognition-ner","task_name":"Named Entity Recognition (NER)"},{"task_slug":"natural-language-inference","task_name":"Natural Language Inference"},{"task_slug":"paraphrase-identification","task_name":"Paraphrase Identification"},{"task_slug":"semantic-similarity","task_name":"Semantic Similarity"},{"task_slug":"semantic-textual-similarity","task_name":"Semantic Textual Similarity"},{"task_slug":"sentiment-analysis","task_name":"Sentiment Analysis"}],"methods":[{"method_slug":"absolute-position-encodings","method_name":"Absolute Position Encodings"},{"method_slug":"adam","method_name":"Adam"},{"method_slug":"attention","method_name":"Attention"},{"method_slug":"bpe","method_name":"BPE"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"label-smoothing","method_name":"Label Smoothing"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"linear-layer","method_name":"Linear Layer"},{"method_slug":"multi-head-attention","method_name":"Multi-Head Attention"},{"method_slug":"position-wise-feed-forward-layer","method_name":"Position-Wise Feed-Forward Layer"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"transformer","method_name":"Transformer"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/machine-reading-comprehension-on-dream","task":"Machine Reading Comprehension","dataset":"DREAM","model":"ASA + RoBERTa","rank_in_archive_order":1,"of":3,"metrics":{"Accuracy":"69.2"},"uses_additional_data":false},{"leaderboard":"/sota/machine-reading-comprehension-on-dream","task":"Machine Reading Comprehension","dataset":"DREAM","model":"ASA + BERT-base","rank_in_archive_order":3,"of":3,"metrics":{"Accuracy":"64.3"},"uses_additional_data":false},{"leaderboard":"/sota/named-entity-recognition-on-wnut-2017","task":"Named Entity Recognition (NER)","dataset":"WNUT 2017","model":"ASA + RoBERTa","rank_in_archive_order":8,"of":23,"metrics":{"F1":"57.3"},"uses_additional_data":false},{"leaderboard":"/sota/named-entity-recognition-on-wnut-2017","task":"Named Entity Recognition (NER)","dataset":"WNUT 2017","model":"ASA + BERT-base","rank_in_archive_order":17,"of":23,"metrics":{"F1":"49.8"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-multinli","task":"Natural Language Inference","dataset":"MultiNLI","model":"ASA + RoBERTa","rank_in_archive_order":18,"of":67,"metrics":{"Matched":"88"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-multinli","task":"Natural Language Inference","dataset":"MultiNLI","model":"ASA + BERT-base","rank_in_archive_order":30,"of":67,"metrics":{"Matched":"85"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-qnli","task":"Natural Language Inference","dataset":"QNLI","model":"ASA + RoBERTa","rank_in_archive_order":20,"of":43,"metrics":{"Accuracy":"93.6%"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-qnli","task":"Natural Language Inference","dataset":"QNLI","model":"ASA + BERT-base","rank_in_archive_order":28,"of":43,"metrics":{"Accuracy":"91.4%"},"uses_additional_data":false},{"leaderboard":"/sota/paraphrase-identification-on-quora-question","task":"Paraphrase Identification","dataset":"Quora Question Pairs","model":"ASA + RoBERTa","rank_in_archive_order":8,"of":31,"metrics":{"F1":"73.7"},"uses_additional_data":false},{"leaderboard":"/sota/paraphrase-identification-on-quora-question","task":"Paraphrase Identification","dataset":"Quora Question Pairs","model":"ASA + BERT-base","rank_in_archive_order":11,"of":31,"metrics":{"F1":"72.3"},"uses_additional_data":false},{"leaderboard":"/sota/semantic-textual-similarity-on-sts-benchmark","task":"Semantic Textual Similarity","dataset":"STS Benchmark","model":"ASA + RoBERTa","rank_in_archive_order":31,"of":66,"metrics":{"Spearman Correlation":"0.892"},"uses_additional_data":false},{"leaderboard":"/sota/semantic-textual-similarity-on-sts-benchmark","task":"Semantic Textual Similarity","dataset":"STS Benchmark","model":"ASA + BERT-base","rank_in_archive_order":44,"of":66,"metrics":{"Spearman Correlation":"0.865"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-2-binary","task":"Sentiment Analysis","dataset":"SST-2 Binary classification","model":"ASA + RoBERTa","rank_in_archive_order":17,"of":87,"metrics":{"Accuracy":"96.3"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-2-binary","task":"Sentiment Analysis","dataset":"SST-2 Binary classification","model":"ASA + BERT-base","rank_in_archive_order":36,"of":87,"metrics":{"Accuracy":"94.1"},"uses_additional_data":false}],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}