{"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/improving-multi-task-deep-neural-networks-via","title":"Improving Multi-Task Deep Neural Networks via Knowledge Distillation for Natural Language Understanding","arxiv_id":"1904.09482","date":"2019-04-20","proceeding":null,"authors":["Xiaodong Liu","Pengcheng He","Weizhu Chen","Jianfeng Gao"],"abstract":"This paper explores the use of knowledge distillation to improve a Multi-Task\nDeep Neural Network (MT-DNN) (Liu et al., 2019) for learning text\nrepresentations across multiple natural language understanding tasks. Although\nensemble learning can improve model performance, serving an ensemble of large\nDNNs such as MT-DNN can be prohibitively expensive. Here we apply the knowledge\ndistillation method (Hinton et al., 2015) in the multi-task learning setting.\nFor each task, we train an ensemble of different MT-DNNs (teacher) that\noutperforms any single model, and then train a single MT-DNN (student) via\nmulti-task learning to \\emph{distill} knowledge from these ensemble teachers.\nWe show that the distilled MT-DNN significantly outperforms the original MT-DNN\non 7 out of 9 GLUE tasks, pushing the GLUE benchmark (single model) to 83.7\\%\n(1.5\\% absolute improvement\\footnote{ Based on the GLUE leaderboard at\nhttps://gluebenchmark.com/leaderboard as of April 1, 2019.}). The code and\npre-trained models will be made publicly available at\nhttps://github.com/namisan/mt-dnn.","url_abs":"http://arxiv.org/abs/1904.09482v1","url_pdf":"http://arxiv.org/pdf/1904.09482v1.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":"improving-multi-task-deep-neural-networks-via","repo_url":"https://github.com/namisan/mt-dnn","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"improving-multi-task-deep-neural-networks-via","repo_url":"https://github.com/chunhuililili/mt_dnn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"improving-multi-task-deep-neural-networks-via","repo_url":"https://github.com/microsoft/MT-DNN","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"ensemble-learning","task_name":"Ensemble Learning"},{"task_slug":"knowledge-distillation","task_name":"Knowledge Distillation"},{"task_slug":"multi-task-learning","task_name":"Multi-Task Learning"},{"task_slug":"natural-language-inference","task_name":"Natural Language Inference"},{"task_slug":"natural-language-understanding","task_name":"Natural Language Understanding"},{"task_slug":"semantic-textual-similarity","task_name":"Semantic Textual Similarity"},{"task_slug":"sentiment-analysis","task_name":"Sentiment Analysis"}],"methods":[{"method_slug":"knowledge-distillation","method_name":"Knowledge Distillation"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/natural-language-inference-on-multinli","task":"Natural Language Inference","dataset":"MultiNLI","model":"MT-DNN-ensemble","rank_in_archive_order":19,"of":67,"metrics":{"Matched":"87.9","Mismatched":"87.4"},"uses_additional_data":false},{"leaderboard":"/sota/semantic-textual-similarity-on-senteval","task":"Semantic Textual Similarity","dataset":"SentEval","model":"MT-DNN-ensemble","rank_in_archive_order":5,"of":6,"metrics":{"MRPC":"92.7/90.3","SICK-E":"-","SICK-R":"-","STS":"91.1/90.7*"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-2-binary","task":"Sentiment Analysis","dataset":"SST-2 Binary classification","model":"MT-DNN-ensemble","rank_in_archive_order":15,"of":87,"metrics":{"Accuracy":"96.5"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1904.09482","atlas_url":"https://app.syntology.ai/?focus=1904.09482","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}