Hybrid Code Networks: Practical and Efficient End-to-End Dialog Control with Supervised and Reinforcement Learning

Hybrid Code Networks: Practical and Efficient End-to-End Dialog Control with Supervised and Reinforcement Learning. Williams, J. D., Asadi, K., & Zweig, G.

Paper abstract bibtex

End-to-end learning of recurrent neural networks (RNNs) is an attractive solution for dialog systems; however, current techniques are data-intensive and require thousands of dialogs to learn simple behaviors. We introduce Hybrid Code Networks (HCNs), which combine an RNN with domain-specific knowledge encoded as software and system action templates. Compared to existing end-to-end approaches, HCNs considerably reduce the amount of training data required, while retaining the key benefit of inferring a latent representation of dialog state. In addition, HCNs can be optimized with supervised learning, reinforcement learning, or a mixture of both. HCNs attain state-of-the-art performance on the bAbI dialog dataset, and outperform two commercially deployed customer-facing dialog systems.

@article{williamsHybridCodeNetworks2017,
  archivePrefix = {arXiv},
  eprinttype = {arxiv},
  eprint = {1702.03274},
  primaryClass = {cs},
  title = {Hybrid {{Code Networks}}: Practical and Efficient End-to-End Dialog Control with Supervised and Reinforcement Learning},
  url = {http://arxiv.org/abs/1702.03274},
  shorttitle = {Hybrid {{Code Networks}}},
  abstract = {End-to-end learning of recurrent neural networks (RNNs) is an attractive solution for dialog systems; however, current techniques are data-intensive and require thousands of dialogs to learn simple behaviors. We introduce Hybrid Code Networks (HCNs), which combine an RNN with domain-specific knowledge encoded as software and system action templates. Compared to existing end-to-end approaches, HCNs considerably reduce the amount of training data required, while retaining the key benefit of inferring a latent representation of dialog state. In addition, HCNs can be optimized with supervised learning, reinforcement learning, or a mixture of both. HCNs attain state-of-the-art performance on the bAbI dialog dataset, and outperform two commercially deployed customer-facing dialog systems.},
  urldate = {2019-02-01},
  date = {2017-02-10},
  keywords = {Computer Science - Artificial Intelligence,Computer Science - Computation and Language},
  author = {Williams, Jason D. and Asadi, Kavosh and Zweig, Geoffrey},
  file = {/home/dimitri/Nextcloud/Zotero/storage/D2SFF2BX/Williams et al. - 2017 - Hybrid Code Networks practical and efficient end-.pdf;/home/dimitri/Nextcloud/Zotero/storage/M4UT656D/1702.html}
}

Downloads: 0

{"_id":"sTd4ffALrm38Ey4vB","bibbaseid":"williams-asadi-zweig-hybridcodenetworkspracticalandefficientendtoenddialogcontrolwithsupervisedandreinforcementlearning","authorIDs":[],"author_short":["Williams, J. D.","Asadi, K.","Zweig, G."],"bibdata":{"bibtype":"article","type":"article","archiveprefix":"arXiv","eprinttype":"arxiv","eprint":"1702.03274","primaryclass":"cs","title":"Hybrid Code Networks: Practical and Efficient End-to-End Dialog Control with Supervised and Reinforcement Learning","url":"http://arxiv.org/abs/1702.03274","shorttitle":"Hybrid Code Networks","abstract":"End-to-end learning of recurrent neural networks (RNNs) is an attractive solution for dialog systems; however, current techniques are data-intensive and require thousands of dialogs to learn simple behaviors. We introduce Hybrid Code Networks (HCNs), which combine an RNN with domain-specific knowledge encoded as software and system action templates. Compared to existing end-to-end approaches, HCNs considerably reduce the amount of training data required, while retaining the key benefit of inferring a latent representation of dialog state. In addition, HCNs can be optimized with supervised learning, reinforcement learning, or a mixture of both. HCNs attain state-of-the-art performance on the bAbI dialog dataset, and outperform two commercially deployed customer-facing dialog systems.","urldate":"2019-02-01","date":"2017-02-10","keywords":"Computer Science - Artificial Intelligence,Computer Science - Computation and Language","author":[{"propositions":[],"lastnames":["Williams"],"firstnames":["Jason","D."],"suffixes":[]},{"propositions":[],"lastnames":["Asadi"],"firstnames":["Kavosh"],"suffixes":[]},{"propositions":[],"lastnames":["Zweig"],"firstnames":["Geoffrey"],"suffixes":[]}],"file":"/home/dimitri/Nextcloud/Zotero/storage/D2SFF2BX/Williams et al. - 2017 - Hybrid Code Networks practical and efficient end-.pdf;/home/dimitri/Nextcloud/Zotero/storage/M4UT656D/1702.html","bibtex":"@article{williamsHybridCodeNetworks2017,\n archivePrefix = {arXiv},\n eprinttype = {arxiv},\n eprint = {1702.03274},\n primaryClass = {cs},\n title = {Hybrid {{Code Networks}}: Practical and Efficient End-to-End Dialog Control with Supervised and Reinforcement Learning},\n url = {http://arxiv.org/abs/1702.03274},\n shorttitle = {Hybrid {{Code Networks}}},\n abstract = {End-to-end learning of recurrent neural networks (RNNs) is an attractive solution for dialog systems; however, current techniques are data-intensive and require thousands of dialogs to learn simple behaviors. We introduce Hybrid Code Networks (HCNs), which combine an RNN with domain-specific knowledge encoded as software and system action templates. Compared to existing end-to-end approaches, HCNs considerably reduce the amount of training data required, while retaining the key benefit of inferring a latent representation of dialog state. In addition, HCNs can be optimized with supervised learning, reinforcement learning, or a mixture of both. HCNs attain state-of-the-art performance on the bAbI dialog dataset, and outperform two commercially deployed customer-facing dialog systems.},\n urldate = {2019-02-01},\n date = {2017-02-10},\n keywords = {Computer Science - Artificial Intelligence,Computer Science - Computation and Language},\n author = {Williams, Jason D. and Asadi, Kavosh and Zweig, Geoffrey},\n file = {/home/dimitri/Nextcloud/Zotero/storage/D2SFF2BX/Williams et al. - 2017 - Hybrid Code Networks practical and efficient end-.pdf;/home/dimitri/Nextcloud/Zotero/storage/M4UT656D/1702.html}\n}\n\n","author_short":["Williams, J. D.","Asadi, K.","Zweig, G."],"key":"williamsHybridCodeNetworks2017","id":"williamsHybridCodeNetworks2017","bibbaseid":"williams-asadi-zweig-hybridcodenetworkspracticalandefficientendtoenddialogcontrolwithsupervisedandreinforcementlearning","role":"author","urls":{"Paper":"http://arxiv.org/abs/1702.03274"},"keyword":["Computer Science - Artificial Intelligence","Computer Science - Computation and Language"],"downloads":0},"bibtype":"article","biburl":"https://raw.githubusercontent.com/dlozeve/newblog/master/bib/all.bib","creationDate":"2020-01-08T20:39:39.244Z","downloads":0,"keywords":["computer science - artificial intelligence","computer science - computation and language"],"search_terms":["hybrid","code","networks","practical","efficient","end","end","dialog","control","supervised","reinforcement","learning","williams","asadi","zweig"],"title":"Hybrid Code Networks: Practical and Efficient End-to-End Dialog Control with Supervised and Reinforcement Learning","year":null,"dataSources":["3XqdvqRE7zuX4cm8m"]}