Meta-supervision for Attention Using Counterfactual Estimation

Seungtaek Choi; Haeju Park; Seung won Hwang

doi:10.1007/s41019-020-00119-z

Meta-supervision for Attention Using Counterfactual Estimation

Seungtaek Choi, Haeju Park, Seung won Hwang

College of Computing

Research output: Contribution to journal › Article › peer-review

1 Citation (Scopus)

Abstract

Neural attention mechanism has been used as a form of explanation for model behavior. Users can either passively consume explanation or actively disagree with explanation and then supervise attention into more proper values (attention supervision). Though attention supervision was shown to be effective in some tasks, we find the existing attention supervision is biased, for which we propose to augment counterfactual observations to debias and contribute to accuracy gains. To this end, we propose a counterfactual method to estimate such missing observations and debias the existing supervisions. We validate the effectiveness of our counterfactual supervision on widely adopted image benchmark datasets: CUFED and PEC.

Original language	English
Pages (from-to)	193-204
Number of pages	12
Journal	Data Science and Engineering
Volume	5
Issue number	2
DOIs	https://doi.org/10.1007/s41019-020-00119-z
Publication status	Published - 2020 Jun 1

Bibliographical note

Funding Information:
This research was supported by the MSIT (Ministry of Science and ICT), Korea, under the ITRC (Information Technology Research Center) support program (IITP-2020-2016-0-00464) supervised by the IITP (Institute of Information and Communications Technology Planning and Evaluation). Hwang is a corresponding author.

Publisher Copyright:
© 2020, The Author(s).

All Science Journal Classification (ASJC) codes

Computational Mechanics
Computer Science Applications

Access to Document

10.1007/s41019-020-00119-z

Cite this

@article{84ed28a429e94363b88e61db1a0d3dea,

title = "Meta-supervision for Attention Using Counterfactual Estimation",

abstract = "Neural attention mechanism has been used as a form of explanation for model behavior. Users can either passively consume explanation or actively disagree with explanation and then supervise attention into more proper values (attention supervision). Though attention supervision was shown to be effective in some tasks, we find the existing attention supervision is biased, for which we propose to augment counterfactual observations to debias and contribute to accuracy gains. To this end, we propose a counterfactual method to estimate such missing observations and debias the existing supervisions. We validate the effectiveness of our counterfactual supervision on widely adopted image benchmark datasets: CUFED and PEC.",

author = "Seungtaek Choi and Haeju Park and Hwang, {Seung won}",

note = "Funding Information: This research was supported by the MSIT (Ministry of Science and ICT), Korea, under the ITRC (Information Technology Research Center) support program (IITP-2020-2016-0-00464) supervised by the IITP (Institute of Information and Communications Technology Planning and Evaluation). Hwang is a corresponding author. Publisher Copyright: {\textcopyright} 2020, The Author(s).",

year = "2020",

month = jun,

day = "1",

doi = "10.1007/s41019-020-00119-z",

language = "English",

volume = "5",

pages = "193--204",

journal = "Data Science and Engineering",

issn = "2364-1185",

publisher = "Springer Berlin",

number = "2",

}

TY - JOUR

T1 - Meta-supervision for Attention Using Counterfactual Estimation

AU - Choi, Seungtaek

AU - Park, Haeju

AU - Hwang, Seung won

N1 - Funding Information: This research was supported by the MSIT (Ministry of Science and ICT), Korea, under the ITRC (Information Technology Research Center) support program (IITP-2020-2016-0-00464) supervised by the IITP (Institute of Information and Communications Technology Planning and Evaluation). Hwang is a corresponding author. Publisher Copyright: © 2020, The Author(s).

PY - 2020/6/1

Y1 - 2020/6/1

N2 - Neural attention mechanism has been used as a form of explanation for model behavior. Users can either passively consume explanation or actively disagree with explanation and then supervise attention into more proper values (attention supervision). Though attention supervision was shown to be effective in some tasks, we find the existing attention supervision is biased, for which we propose to augment counterfactual observations to debias and contribute to accuracy gains. To this end, we propose a counterfactual method to estimate such missing observations and debias the existing supervisions. We validate the effectiveness of our counterfactual supervision on widely adopted image benchmark datasets: CUFED and PEC.

AB - Neural attention mechanism has been used as a form of explanation for model behavior. Users can either passively consume explanation or actively disagree with explanation and then supervise attention into more proper values (attention supervision). Though attention supervision was shown to be effective in some tasks, we find the existing attention supervision is biased, for which we propose to augment counterfactual observations to debias and contribute to accuracy gains. To this end, we propose a counterfactual method to estimate such missing observations and debias the existing supervisions. We validate the effectiveness of our counterfactual supervision on widely adopted image benchmark datasets: CUFED and PEC.

UR - http://www.scopus.com/inward/record.url?scp=85084977698&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85084977698&partnerID=8YFLogxK

U2 - 10.1007/s41019-020-00119-z

DO - 10.1007/s41019-020-00119-z

M3 - Article

AN - SCOPUS:85084977698

SN - 2364-1185

VL - 5

SP - 193

EP - 204

JO - Data Science and Engineering

JF - Data Science and Engineering

IS - 2

ER -

Meta-supervision for Attention Using Counterfactual Estimation

Abstract

Bibliographical note

All Science Journal Classification (ASJC) codes

Access to Document

Other files and links

Fingerprint

Cite this