@inproceedings{arora-etal-2019-knowledge,
title = "On Knowledge distillation from complex networks for response prediction",
author = "Arora, Siddhartha and
Khapra, Mitesh M. and
Ramaswamy, Harish G.",
editor = "Burstein, Jill and
Doran, Christy and
Solorio, Thamar",
booktitle = "Proceedings of the 2019 Conference of the North {A}merican Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers)",
month = jun,
year = "2019",
address = "Minneapolis, Minnesota",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/N19-1382",
doi = "10.18653/v1/N19-1382",
pages = "3813--3822",
abstract = "Recent advances in Question Answering have lead to the development of very complex models which compute rich representations for query and documents by capturing all pairwise interactions between query and document words. This makes these models expensive in space and time, and in practice one has to restrict the length of the documents that can be fed to these models. Such models have also been recently employed for the task of predicting dialog responses from available background documents (e.g., Holl-E dataset). However, here the documents are longer, thereby rendering these complex models infeasible except in select restricted settings. In order to overcome this, we use standard simple models which do not capture all pairwise interactions, but learn to emulate certain characteristics of a complex teacher network. Specifically, we first investigate the conicity of representations learned by a complex model and observe that it is significantly lower than that of simpler models. Based on this insight, we modify the simple architecture to mimic this characteristic. We go further by using knowledge distillation approaches, where the simple model acts as a student and learns to match the output from the complex teacher network. We experiment with the Holl-E dialog data set and show that by mimicking characteristics and matching outputs from a teacher, even a simple network can give improved performance.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="arora-etal-2019-knowledge">
<titleInfo>
<title>On Knowledge distillation from complex networks for response prediction</title>
</titleInfo>
<name type="personal">
<namePart type="given">Siddhartha</namePart>
<namePart type="family">Arora</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mitesh</namePart>
<namePart type="given">M</namePart>
<namePart type="family">Khapra</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Harish</namePart>
<namePart type="given">G</namePart>
<namePart type="family">Ramaswamy</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2019-06</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jill</namePart>
<namePart type="family">Burstein</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christy</namePart>
<namePart type="family">Doran</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Thamar</namePart>
<namePart type="family">Solorio</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Minneapolis, Minnesota</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Recent advances in Question Answering have lead to the development of very complex models which compute rich representations for query and documents by capturing all pairwise interactions between query and document words. This makes these models expensive in space and time, and in practice one has to restrict the length of the documents that can be fed to these models. Such models have also been recently employed for the task of predicting dialog responses from available background documents (e.g., Holl-E dataset). However, here the documents are longer, thereby rendering these complex models infeasible except in select restricted settings. In order to overcome this, we use standard simple models which do not capture all pairwise interactions, but learn to emulate certain characteristics of a complex teacher network. Specifically, we first investigate the conicity of representations learned by a complex model and observe that it is significantly lower than that of simpler models. Based on this insight, we modify the simple architecture to mimic this characteristic. We go further by using knowledge distillation approaches, where the simple model acts as a student and learns to match the output from the complex teacher network. We experiment with the Holl-E dialog data set and show that by mimicking characteristics and matching outputs from a teacher, even a simple network can give improved performance.</abstract>
<identifier type="citekey">arora-etal-2019-knowledge</identifier>
<identifier type="doi">10.18653/v1/N19-1382</identifier>
<location>
<url>https://aclanthology.org/N19-1382</url>
</location>
<part>
<date>2019-06</date>
<extent unit="page">
<start>3813</start>
<end>3822</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T On Knowledge distillation from complex networks for response prediction
%A Arora, Siddhartha
%A Khapra, Mitesh M.
%A Ramaswamy, Harish G.
%Y Burstein, Jill
%Y Doran, Christy
%Y Solorio, Thamar
%S Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers)
%D 2019
%8 June
%I Association for Computational Linguistics
%C Minneapolis, Minnesota
%F arora-etal-2019-knowledge
%X Recent advances in Question Answering have lead to the development of very complex models which compute rich representations for query and documents by capturing all pairwise interactions between query and document words. This makes these models expensive in space and time, and in practice one has to restrict the length of the documents that can be fed to these models. Such models have also been recently employed for the task of predicting dialog responses from available background documents (e.g., Holl-E dataset). However, here the documents are longer, thereby rendering these complex models infeasible except in select restricted settings. In order to overcome this, we use standard simple models which do not capture all pairwise interactions, but learn to emulate certain characteristics of a complex teacher network. Specifically, we first investigate the conicity of representations learned by a complex model and observe that it is significantly lower than that of simpler models. Based on this insight, we modify the simple architecture to mimic this characteristic. We go further by using knowledge distillation approaches, where the simple model acts as a student and learns to match the output from the complex teacher network. We experiment with the Holl-E dialog data set and show that by mimicking characteristics and matching outputs from a teacher, even a simple network can give improved performance.
%R 10.18653/v1/N19-1382
%U https://aclanthology.org/N19-1382
%U https://doi.org/10.18653/v1/N19-1382
%P 3813-3822
Markdown (Informal)
[On Knowledge distillation from complex networks for response prediction](https://aclanthology.org/N19-1382) (Arora et al., NAACL 2019)
ACL
- Siddhartha Arora, Mitesh M. Khapra, and Harish G. Ramaswamy. 2019. On Knowledge distillation from complex networks for response prediction. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 3813–3822, Minneapolis, Minnesota. Association for Computational Linguistics.