{"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/learning-universal-sentence-representations","title":"Learning Universal Sentence Representations with Mean-Max Attention Autoencoder","arxiv_id":"1809.06590","date":"2018-09-18","proceeding":"EMNLP 2018 10","authors":["Minghua Zhang","Yunfang Wu","Weikang Li","Wei Li"],"abstract":"In order to learn universal sentence representations, previous methods focus\non complex recurrent neural networks or supervised learning. In this paper, we\npropose a mean-max attention autoencoder (mean-max AAE) within the\nencoder-decoder framework. Our autoencoder rely entirely on the MultiHead\nself-attention mechanism to reconstruct the input sequence. In the encoding we\npropose a mean-max strategy that applies both mean and max pooling operations\nover the hidden vectors to capture diverse information of the input. To enable\nthe information to steer the reconstruction process dynamically, the decoder\nperforms attention over the mean-max representation. By training our model on a\nlarge collection of unlabelled data, we obtain high-quality representations of\nsentences. Experimental results on a broad range of 10 transfer tasks\ndemonstrate that our model outperforms the state-of-the-art unsupervised single\nmethods, including the classical skip-thoughts and the advanced\nskip-thoughts+LN model. Furthermore, compared with the traditional recurrent\nneural network, our mean-max AAE greatly reduce the training time.","url_abs":"http://arxiv.org/abs/1809.06590v1","url_pdf":"http://arxiv.org/pdf/1809.06590v1.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":"learning-universal-sentence-representations","repo_url":"https://github.com/Zminghua/SentEncoding","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":null}],"tasks":[{"task_slug":"decoder","task_name":"Decoder"},{"task_slug":"sentence","task_name":"Sentence"}],"methods":[{"method_slug":"max-pooling","method_name":"Max Pooling"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}