{"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/double-path-networks-for-sequence-to-sequence","title":"Double Path Networks for Sequence to Sequence Learning","arxiv_id":"1806.04856","date":"2018-06-13","proceeding":"COLING 2018 8","authors":["Kaitao Song","Xu Tan","Di He","Jianfeng Lu","Tao Qin","Tie-Yan Liu"],"abstract":"Encoder-decoder based Sequence to Sequence learning (S2S) has made remarkable\nprogress in recent years. Different network architectures have been used in the\nencoder/decoder. Among them, Convolutional Neural Networks (CNN) and Self\nAttention Networks (SAN) are the prominent ones. The two architectures achieve\nsimilar performances but use very different ways to encode and decode context:\nCNN use convolutional layers to focus on the local connectivity of the\nsequence, while SAN uses self-attention layers to focus on global semantics. In\nthis work we propose Double Path Networks for Sequence to Sequence learning\n(DPN-S2S), which leverage the advantages of both models by using double path\ninformation fusion. During the encoding step, we develop a double path\narchitecture to maintain the information coming from different paths with\nconvolutional layers and self-attention layers separately. To effectively use\nthe encoded context, we develop a cross attention module with gating and use it\nto automatically pick up the information needed during the decoding step. By\ndeeply integrating the two paths with cross attention, both types of\ninformation are combined and well exploited. Experiments show that our proposed\nmethod can significantly improve the performance of sequence to sequence\nlearning over state-of-the-art systems.","url_abs":"http://arxiv.org/abs/1806.04856v2","url_pdf":"http://arxiv.org/pdf/1806.04856v2.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":"double-path-networks-for-sequence-to-sequence","repo_url":"https://github.com/StillKeepTry/Transformer-PyTorch","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"decoder","task_name":"Decoder"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1806.04856","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}