Information: Join today and make your research activities more affordable! Technical workshop participation fees and annual registration fees are available at member rates.
Notice: [Important] Announcement of Changes to Registration Fee Payment and Manuscript Upload Procedures for IEICE Technical Meetings
IEICE Technical Committee Submission System
Conference Paper's Information		 Online Proceedings
[Sign in]
Tech. Rep. Archives
 Go Top Page Go Previous   [Japanese] / [English] 

Paper Abstract and Keywords
Presentation 2020-10-30 17:35
Hyperbolic Space Embedding for Open Set Recognition
Shota Tatarai (Senshu Univ.), Yuta Ashihara (Nihon Univ/Glia Computing Co.,Ltd.), Kenji Aoki (Glia Computing Co.,Ltd.), Masahiko Osaawa (Nihon Univ./Senshu Univ.) NC2020-27
Abstract (in Japanese) (See Japanese page) 
(in English) Many of deep learning algorithms perform well when the training and testing data are sampled from the
same class space. However, in an uncontrolled
environment, such as a real-world scenario, models have to handle
unwanted or unknown inputs. This problem is a crucial component of real-world applications.
In this paper, we propose a new method that uses the Poincare ball model for embedding features
to reject unknown inputs as an unknown class. The proposed method adopts Triplet Loss
which employs RiemaniannSGD to embed the features into the hyperbolic space.
In our experiments, we checked the performance of the proposed method
through compared it with existing method, using CIFAR-10 as a training dataset
and SVHN as an unknown dataset. Our method achieved 82.15% accuracy. Furthermore,
we found that the unknown inputs have a specific direction, by visualizing the Poincare ball.
Through this study, we summarize that our approach has
the possibility of developing a hyperbolic space approach for handling unknown inputs.
Keyword (in Japanese) (See Japanese page) 
(in English) deep learning / metric learning / unknown detection / hyperbolic space / / / /  
Reference Info. IEICE Tech. Rep., vol. 120, no. 216, NC2020-27, pp. 100-105, Oct. 2020.
Paper # NC2020-27 
Date of Issue 2020-10-22 (NC) 
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 NC2020-27

Conference Information
Committee MBE NC NLP CAS  
Conference Date 2020-10-29 - 2020-10-30 
Place (in Japanese) (See Japanese page) 
Place (in English) Online 
Topics (in Japanese) (See Japanese page) 
Topics (in English) ME，NC，CAS，NLP 
Paper Information
Registration To NC 
Conference Code 2020-10-MBE-NC-NLP-CAS 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Hyperbolic Space Embedding for Open Set Recognition 
Sub Title (in English)  
Keyword(1) deep learning  
Keyword(2) metric learning  
Keyword(3) unknown detection  
Keyword(4) hyperbolic space  
Keyword(5)  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Shota Tatarai  
1st Author's Affiliation Senshu University (Senshu Univ.)
2nd Author's Name Yuta Ashihara  
2nd Author's Affiliation Nihon University/Glia Computing Co.,Ltd. (Nihon Univ/Glia Computing Co.,Ltd.)
3rd Author's Name Kenji Aoki  
3rd Author's Affiliation Glia Computing Co.,Ltd. (Glia Computing Co.,Ltd.)
4th Author's Name Masahiko Osaawa  
4th Author's Affiliation Nihon University/Senshu University (Nihon Univ./Senshu Univ.)
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 ()
21st Author's Name  
21st Author's Affiliation ()
22nd Author's Name  
22nd Author's Affiliation ()
23rd Author's Name  
23rd Author's Affiliation ()
24th Author's Name  
24th Author's Affiliation ()
25th Author's Name  
25th Author's Affiliation ()
26th Author's Name / /
26th Author's Affiliation ()
()
27th Author's Name / /
27th Author's Affiliation ()
()
28th Author's Name / /
28th Author's Affiliation ()
()
29th Author's Name / /
29th Author's Affiliation ()
()
30th Author's Name / /
30th Author's Affiliation ()
()
31st Author's Name / /
31st Author's Affiliation ()
()
32nd Author's Name / /
32nd Author's Affiliation ()
()
33rd Author's Name / /
33rd Author's Affiliation ()
()
34th Author's Name / /
34th Author's Affiliation ()
()
35th Author's Name / /
35th Author's Affiliation ()
()
36th Author's Name / /
36th Author's Affiliation ()
()
Speaker Author-1 
Date Time 2020-10-30 17:35:00 
Presentation Time 25 minutes 
Registration for NC 
Paper # NC2020-27 
Volume (vol) vol.120 
Number (no) no.216 
Page pp.100-105 
#Pages 6 
Date of Issue 2020-10-22 (NC) 

[Return to Top Page]

[Return to IEICE Web Page]
The Institute of Electronics, Information and Communication Engineers (IEICE), Japan