| Paper Abstract and Keywords |
| Presentation |
2026-03-25 11:20
Acceleration of deep equilibrium models based on Banach fixed point theorem Naoki Sato, Hideaki Iiduka (Meiji Univ.) IBISML2025-50 |
| Abstract |
(in Japanese) |
(See Japanese page) |
| (in English) |
The Deep Equilibrium Model (DEQ) achieves infinitely deep network representations without stacking layers by exploring fixed points of layer transformations in neural networks. Despite requiring significantly less memory than conventional models, DEQ demonstrates performance comparable to state-of-the-art models in numerous large-scale numerical experiments. However, it faces the challenge of requiring enormous computational time for both training and inference compared to conventional models, as it performs fixed-point iterations with no convergence guarantee each time it receives input. This paper introduces an approach that improves fixed-point convergence and consequently reduces computational time by reconstructing the model architecture to guarantee fixed-point convergence.
Our proposed Lipschitz MDEQ for image classification theoretically guarantees fixed-point convergence in both the forward and backward passes through hyperparameter tuning. Numerical experiments on the CIFAR-10 dataset achieved up to 4.75x faster speedup at the cost of a slight accuracy drop. |
| Keyword |
(in Japanese) |
(See Japanese page) |
| (in English) |
Deep equilibrium models / Banach fixed point theorem / Contractive mapping / / / / / |
| Reference Info. |
IEICE Tech. Rep., vol. 125, no. 425, IBISML2025-50, pp. 67-74, March 2026. |
| Paper # |
IBISML2025-50 |
| Date of Issue |
2026-03-17 (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 |
IBISML2025-50 |
| Conference Information |
| Committee |
PRMU IPSJ-CVIM IBISML ITE-SIP |
| Conference Date |
2026-03-24 - 2026-03-25 |
| Place (in Japanese) |
(See Japanese page) |
| Place (in English) |
|
| Topics (in Japanese) |
(See Japanese page) |
| Topics (in English) |
|
| Paper Information |
| Registration To |
IBISML |
| Conference Code |
2026-03-PRMU-CVIM-IBISML-SIP |
| Language |
Japanese |
| Title (in Japanese) |
(See Japanese page) |
| Sub Title (in Japanese) |
(See Japanese page) |
| Title (in English) |
Acceleration of deep equilibrium models based on Banach fixed point theorem |
| Sub Title (in English) |
|
| Keyword(1) |
Deep equilibrium models |
| Keyword(2) |
Banach fixed point theorem |
| Keyword(3) |
Contractive mapping |
| Keyword(4) |
|
| Keyword(5) |
|
| Keyword(6) |
|
| Keyword(7) |
|
| Keyword(8) |
|
| 1st Author's Name |
Naoki Sato |
| 1st Author's Affiliation |
Meiji University (Meiji Univ.) |
| 2nd Author's Name |
Hideaki Iiduka |
| 2nd Author's Affiliation |
Meiji University (Meiji Univ.) |
| 3rd Author's Name |
|
| 3rd Author's Affiliation |
() |
| 4th Author's Name |
|
| 4th Author's Affiliation |
() |
| 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 |
2026-03-25 11:20:00 |
| Presentation Time |
20 minutes |
| Registration for |
IBISML |
| Paper # |
IBISML2025-50 |
| Volume (vol) |
vol.125 |
| Number (no) |
no.425 |
| Page |
pp.67-74 |
| #Pages |
8 |
| Date of Issue |
2026-03-17 (IBISML) |