Natori, Naotake

写真a

Affiliation

Research Centers and Institutes, Keio University Global Research Institute ( Mita )

Position

Project Professor (Non-tenured)
Page Top ▲
 

Books 【 Display / hide 】

  • 機械学習デザインパターン : データ準備、モデル構築、MLOpsの実践上の問題と解決

    鷲崎 弘宜, 竹内 広宜, 名取 直毅, 吉岡 信和, オライリー・ジャパン,オーム社 (発売), 2021.10,  Page： xxi, 387p

Page Top ▲

Papers 【 Display / hide 】

  • Transfer multi-source knowledge via scale-aware online domain adaptation in depth estimation for autonomous driving

    Phan Thi Huyen Thanh, Minh Quan Viet Bui, Duc Dung Nguyen, Tran Vu Pham, Truong Vinh Truong Duy, Naotake Natori

    Image and Vision Computing (IMAVIS)  2024

    Last author, Accepted

  • A Framework for Developing Reliable Machine Learning Systems and its Application

    鷲崎 弘宜, Husen Jati, Runpakprakun Jomphon, Guan Shiyang, 吉岡 信和, 名取 直毅, Duy Truong

    Proceedings of the Annual Conference of the Japanese Society for Artificial Intelligence (JSAI) (The Japanese Society for Artificial Intelligence)  38 2024

    Accepted,  ISSN  2758-7347

    　View Summary

    We describe M3S (Multi-view Modeling framework for ML systems), a framework that integrates multi-view and consistent analytical and design modeling and machine learning model workflow pipelines for the continuous development and operation of highly reliable machine learning systems, and its application example in the mobility domain.

  • Systematic design of artificial deep neural networks based on scaling laws in signal propagation

    玉井 敬一, 大久保 毅, ズイ チュオン ビン チュオン, 名取 直毅, 藤堂 眞治

    Proceedings of the Annual Conference of the Japanese Society for Artificial Intelligence (JSAI) (The Japanese Society for Artificial Intelligence)  38 2024

    Accepted,  ISSN  2758-7347

    　View Summary

    For more environmentally sustainable development of deep learning (DL) technologies, computational burden for tuning DL architectures should be reduced. This calls for more systematic strategies for finding an optimal set of hyperparameters which achieves a good balance between training speed and generalization performance. As a preliminary step toward this goal, we address the problem of how to tune fully-connected feedforward perceptrons in the so-called ``kernel regime'' in a systematic manner. By combining the existing theoretical tools, such as the Neural Tangent Kernel (NTK), and the analogy of the signal propagation dynamics with absorbing phase transitions, we conduct thorough analysis of the training dynamics of the neural network, including the case with finite depth. As a result, a simple strategy for optimally tuning the initialization hyperparameters and the depth is proposed.

  • Unsupervised Moving Object Segmentation and Ego-Velocity Prediction for Autonomous Vehicles

    Ul Haq Israr, Phan Thi Huyen Thanh, Yuichiro Yoshimura, Truong Vinh Truong Duy, Naotake Natori

    Proceedings of the Annual Conference of the Japanese Society for Artificial Intelligence (JSAI) (The Japanese Society for Artificial Intelligence)  38 2024

    Last author, Accepted

    　View Summary

    Motion segmentation in computer vision is a challenging task, particularly in the context of self-driving vehicles where backgrounds are constantly changing. Accurately detecting moving objects is crucial for effective vehicle control. To address this, we propose an innovative approach called Unsupervised Moving Object and Ego-Velocity Prediction (UMVP) specifically designed for autonomous vehicles. UMVP utilizes depth maps predicted from RGB images and trains a motion network using these depth maps and consecutive pairs of RGB frames. Additionally, it predicts the speed of the ego-vehicle by analyzing a pair of images. Our approach is completely unsupervised, eliminating the need for manual annotation or labeled data. We evaluated UMVP on the KITTI dataset, and observed significant improvements in motion segmentation, depth estimation compared to the baseline method. These results highlight the potential of UMVP to enhance motion segmentation in autonomous vehicles.

  • Absorbing phase transitions in artificial deep neural networks

    玉井 敬一, 大久保 毅, ヴィン チュオン ズイ チュオン, 名取 直毅, 藤堂 眞治

    Proceedings of the Annual Conference of JSAI (The Japanese Society for Artificial Intelligence)  37 2023

    Accepted,  ISSN  2758-7347

    　View Summary

    For wider use of deep learning (DL) in society, deeper understanding of fundamental principles underlying various DL architectures is needed so that the users have better control over what they are actually doing with DL technologies. The deeper understanding is also likely to be useful for developing more environmentally friendly learning methodologies. As a preliminary step toward this goal, we study fundamental properties of signal propagations in artificial deep neural networks in this paper. More specifically, we show that there is a strong analogy between the signal propagation process in appropriately initialized fully-connected/convolutional deep neural networks and the dynamics of the so-called "absorbing phase transitions (APTs)" which can be found in some physical systems driven far away from equilibrium. We discuss, with numerical results on the signal propagation process, how these neural networks can be placed in a context of the theory of APTs and what theoretical/practical implication can be gained beyond the well-known mean-field theory.

display all >>

Page Top ▲

Reviews, Commentaries, etc. 【 Display / hide 】

  • OS-17「ひと中心の未来社会とAI」

    名取 直毅, 梶 大介, 廣瀬 正明, 河村 芳海, 梶 洋隆, 城殿 清澄

    人工知能 39 ( 6 )  2024

  • OS-20「ひと中心の未来社会とAI」

    名取 直毅, 梶 大介, 廣瀬 正明, 河村 芳海, 梶 洋隆, 城殿 清澄

    人工知能 38 ( 6 )  2023

  • 会議報告：The 14th International Conference on Knowledge, Information and Creativity Support Systems (KICSS 2019)

    Truong Duy, 名取 直毅

    人工知能 (一般社団法人 人工知能学会)  35 ( 2 ) 309 - 310 2020

    Last author,  ISSN  2188-2266

Page Top ▲

Presentations 【 Display / hide 】

display all >>

Page Top ▲

Intellectual Property Rights, etc. 【 Display / hide 】

  • 情報処理装置

    Date applied： 特願2023-128353  2023.08 

    Date announced： 特開2025-024324  2025.02 

    Patent

  • 情報処理装置

    Date applied： 特願2022-158116  2022.09 

    Date announced： 特開2024-051786  2024.04 

    Patent

  • 異常検出装置、異常検出プログラム、および異常検出システム

    Date applied： 特願2020-154964  2020.09 

    Date announced： 特開2022-048904  2022.03 

    Date issued： 特許第7501264号 

    Date registered： 2024.06

    Patent

  • 異常検出装置、異常検出プログラム、および異常検出システム

    Date applied： 特願2020-154964  2020.09 

    Date announced： 特開2022-048904  2022.03 

    Patent

  • 紙葉類処理装置、紙葉類処理方法、および、紙葉類処理プログラム

    Date applied： 特願2017-055056  2017.03 

    Date announced： 特開2018-156605  2018.10 

    Patent

display all >>

Page Top ▲

Awards 【 Display / hide 】

  • 2022 Best Paper Award

    Hironori Washizaki, Foutse Khomh, Yann-Gaël Guéhéneuc, Hironori Takeuchi, Naotake Natori, Takuo Doi, Satoshi Okuda, 2023, IEEE Computer Society, Software-Engineering Design Patterns for Machine Learning Applications

    Type of Award： International academic award (Japan or overseas)

  • MIRU2021優秀賞

    池川 慎一, 齊院 龍二, 澤田 好秀, 名取 直毅, 2021, CVIM/PRMU画像の認識・理解シンポジウム (MIRU), 正規化と Pre-Activation モジュールを用いた深層スパイキングニューラルネットワーク

    Type of Award： Award from Japanese society, conference, symposium, etc.

  • 1996年優秀論文発表賞

    Naotake Natori, Kazuo Nishimura, 1996, 一般社団法人 電気学会, A Practical Neural Network for Handwritten Character Recognition Based on dynamics-Based Active Learning and Self-Organization of Feedback

Page Top ▲

Other 【 Display / hide 】

  • 人工知能学会全国大会OS-17「ひと中心の未来社会とAI」オーガナイザ

    　View Details

    共同でオーガナイザを務めた。

  • 人工知能学会全国大会OS-20「ひと中心の未来社会とAI」オーガナイザ

    　View Details

    共同でオーガナイザを務めた。

Page Top ▲