Paper Abstract and Keywords |
Presentation |
2020-10-22 14:50
Suppressing explanations with irrelevant concepts in deep learning Munemasa Tomohiro (Tsukuba Univ), Fukuchi Kazuto, Akimoto Yohei, Sakuma Jun (Tsukuba Univ/Riken AIP) IBISML2020-32 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
TCAV [1], which is an explanation method using a concept that humans easily understand for deep learning models, concept based and post-hoc explanation method. TCAV is a method based on importance of concept given in advance (e.g. roundness, red) in a certain class. In this study, we experimentally confirmed that TCAV frequently causes false positives that the concepts unrelated to the class is important. Furthermore we clarified the cause of false positives is a sample bias used to generate the concept. We experimentally confirmed that false positives counld be suppressed by eliminating the cause. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
XAI / Post-hoc explanation / Concept / False Positive / / / / |
Reference Info. |
IEICE Tech. Rep., vol. 120, no. 195, IBISML2020-32, pp. 61-68, Oct. 2020. |
Paper # |
IBISML2020-32 |
Date of Issue |
2020-10-13 (IBISML) |
ISSN |
Online edition: ISSN 2432-6380 |
Copyright and reproduction |
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
Download PDF |
IBISML2020-32 |
Conference Information |
Committee |
IBISML |
Conference Date |
2020-10-20 - 2020-10-22 |
Place (in Japanese) |
(See Japanese page) |
Place (in English) |
Online |
Topics (in Japanese) |
(See Japanese page) |
Topics (in English) |
Organized Sessions on Frontiers of Machine Learning and General Sessions |
Paper Information |
Registration To |
IBISML |
Conference Code |
2020-10-IBISML |
Language |
Japanese |
Title (in Japanese) |
(See Japanese page) |
Sub Title (in Japanese) |
(See Japanese page) |
Title (in English) |
Suppressing explanations with irrelevant concepts in deep learning |
Sub Title (in English) |
|
Keyword(1) |
XAI |
Keyword(2) |
Post-hoc explanation |
Keyword(3) |
Concept |
Keyword(4) |
False Positive |
Keyword(5) |
|
Keyword(6) |
|
Keyword(7) |
|
Keyword(8) |
|
1st Author's Name |
Munemasa Tomohiro |
1st Author's Affiliation |
Tsukuba University (Tsukuba Univ) |
2nd Author's Name |
Fukuchi Kazuto |
2nd Author's Affiliation |
Tsukuba University /Riken AIP (Tsukuba Univ/Riken AIP) |
3rd Author's Name |
Akimoto Yohei |
3rd Author's Affiliation |
Tsukuba University /Riken AIP (Tsukuba Univ/Riken AIP) |
4th Author's Name |
Sakuma Jun |
4th Author's Affiliation |
Tsukuba University /Riken AIP (Tsukuba Univ/Riken AIP) |
5th Author's Name |
|
5th Author's Affiliation |
() |
6th Author's Name |
|
6th Author's Affiliation |
() |
7th Author's Name |
|
7th Author's Affiliation |
() |
8th Author's Name |
|
8th Author's Affiliation |
() |
9th Author's Name |
|
9th Author's Affiliation |
() |
10th Author's Name |
|
10th Author's Affiliation |
() |
11th Author's Name |
|
11th Author's Affiliation |
() |
12th Author's Name |
|
12th Author's Affiliation |
() |
13th Author's Name |
|
13th Author's Affiliation |
() |
14th Author's Name |
|
14th Author's Affiliation |
() |
15th Author's Name |
|
15th Author's Affiliation |
() |
16th Author's Name |
|
16th Author's Affiliation |
() |
17th Author's Name |
|
17th Author's Affiliation |
() |
18th Author's Name |
|
18th Author's Affiliation |
() |
19th Author's Name |
|
19th Author's Affiliation |
() |
20th Author's Name |
|
20th Author's Affiliation |
() |
Speaker |
Author-1 |
Date Time |
2020-10-22 14:50:00 |
Presentation Time |
25 minutes |
Registration for |
IBISML |
Paper # |
IBISML2020-32 |
Volume (vol) |
vol.120 |
Number (no) |
no.195 |
Page |
pp.61-68 |
#Pages |
8 |
Date of Issue |
2020-10-13 (IBISML) |
|