Research Interests

• Machine Learning

• Deep Learning Theory

• Sequential Probability Ratio Test (SPRT)

• Early Classification of Time Series (ECTS)

• Density Ratio Estimation (DRE)

• Quickest Changepoint Detection (QCD)

• AI Scientist

Career Summary

• 2010-2017    BS, MS, PhD student at Kyoto University, Japan

• 2017-             Researcher at NEC Corporation, Japan

• 2019-2022    Visiting scientist at RIKEN AIP, Japan

• 2023-             Independent researcher

Publications

Peer-reviewed Papers

17. Taiki Miyagawa. A Link between Shock-wave Theory and Symmetry-reduced Stochastic Gradient Descent for Artificial Neural Networks [arXiv TBA]. International Conference on Artificial Neural Networks (ICANN), 2026.

17. Taiki Miyagawa, Akinori F. Ebihara. Accurate Evaluation of Quickest Changepoint Detectors via Non-parametric Survival Analysis [OpenReview] [arXiv] [GitHub]. International Conference on Machine Learning (ICML), 2026.

16.  Junki Mori, Kazuya Kakizaki, Taiki Miyagawa, Jun Sakuma. Differentially Private Synthetic Text Generation for Retrieval-Augmented Generation (RAG) [arXiv]. Annual Meeting of the Association for Computational Linguistics (ACL) Findings, 2026

15. Kosuke Kihara, Junki Mori, Taiki Miyagawa, Akinori F. Ebihara. Rethinking the Backbone in Class Imbalanced Federated Source Free Domain Adaptation: The Utility of Vision Foundation Models [arXiv]. IEEE International Conference on Image Processing (ICIP) Workshop 1: Edge Intelligence: Smart, Efficient, and Scalable Solutions for IoT, Wearables, and Embedded Devices (SEEDS), 2025.

14. Takuma Amada, Kazuya Kakizaki, Taiki Miyagawa, Akinori F. Ebihara, Kaede Shiohara, Toshihiko Yamasaki. Robust Deepfake Detection for Electronic Know Your Customer Systems Using Registered Images [arXiv] [IEEE]. IEEE International Conference on Automatic Face and Gesture Recognition (FG), 2025.

13. Akinori F. Ebihara, Taiki Miyagawa, Kazuyuki Sakurai, Hitoshi Imaoka. Learning the Optimal Stopping for Early Classification within Finite Horizons via Sequential Probability Ratio Test [OpenReview] [arXiv] [Graphical abstract 1/2] [Graphical abstract 2/2]. International Conference on Learning Representations (ICLR), 2025.

12. Junki Mori, Kosuke Kihara, Taiki Miyagawa, Akinori F. Ebihara, Isamu Teranishi, Hisashi Kashima. Federated Source-free Domain Adaptation for Classification: Weighted Cluster Aggregation for Unlabeled Data [arXiv]. IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2024.

11. Yuto Matsunaga, Satoru Momiyama, Batnyam Enkhtaivan, Taiki Miyagawa,  Isamu Teranishi. ConvFL: Efficient Federated Learning of Heterogeneous Model Structures with Converters [IEEE]. IEEE International Conference on Smart Internet of Things (SmartIoT), 2024.

10. Taiki Miyagawa*, Takeru Yokota* (* Equal contribution). Physics-informed Neural Networks for Functional Differential Equations: Cylindrical Approximation and Its Convergence Guarantees [OpenReview] [Presentation video]. Annual Conference on Neural Information Processing Systems (NeurIPS), 2024. 

9. Hideyuki Oiso, Yuto Matsunaga, Kazuya Kakizaki, Taiki Miyagawa. Prompt Tuning for Audio Deepfake Detection: Computationally Efficient Test-time Domain Adaptation with Limited Target Dataset [arXiv]. Interspeech, 2024.

8. Hiroshi Fukui, Taiki Miyagawa,  Yusuke Morishita. Multi-object tracking as attention mechanism [arXiv]. IEEE International conference on image processing (ICIP), 2023.

7. Akinori F. Ebihara, Taiki Miyagawa, Kazuyuki Sakurai, Hitoshi Imaoka. Toward asymptotic optimality: Sequential unsupervised regression of density ratio for early classification [arXiv]. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023.

6. Taiki Miyagawa. Toward equation of motion for deep neural networks: Continuous-time gradient descent and discretization error analysis [OpenReview] [Presentation video]. Advances in Neural Information Processing Systems (NeurIPS), 2022.

5. Azusa Sawada, Taiki Miyagawa, Akinori Ebihara, Shoji Yachida, Toshinori Hosoi. Convolutional neural networks for time-dependent classification of variable-length time series [arXiv]. IEEE International Joint Conference on Neural Networks (IJCNN), 2022. * Oral presentation. *

4. Kazuya Kakizaki, Taiki Miyagawa, Inderjeet Singh, Jun Sakuma. Toward practical adversarial attacks on face verification systems [PDF]. IEEE International Conference of the Biometrics Special Interest Group (BIOSIG), 2021.

3. Taiki Miyagawa, Akinori F. Ebihara. The power of log-sum-exp: Sequential density ratio matrix estimation for speed-accuracy optimization [arXiv] [Presentation video]. International Conference on Machine Learning (ICML), 2021.

2. Akinori F. Ebihara, Taiki Miyagawa, Kazuyuki Sakurai, Hitoshi Imaoka. Sequential density ratio estimation for simultaneous optimization of speed and accuracy [OpenReview] [arXiv]. International Conference on Learning Representations (ICLR), 2021. * Spotlight paper: top 5.57%. * * Exhibited at the Sumitomo Pavilion, Expo 2025 Osaka, Kansai. [My tweet (JP)] *

1. Taiki Miyagawa, Noburo Shiba, Tadashi Takayanagi. Double‐trace deformations and entanglement entropy in AdS [arXiv] [WILEY]. Fortschritte der Physik, 2016

Peer-reviewed Domestic Papers

 1. Taiki Miyagawa,  Akinori F. Ebihara. 統計的早期時系列分類のレビュー：理論的に最適な早押し分類器の実用化 (Review: Statistical Early Classification of Time Series) [URL: participants only]. Symposium on Sensing via Image Information (SSII), 2025. (in Japanese)

Non-peer-reviewed Papers

8. Kosuke Akimoto, Taiki Miyagawa, Masafumi Oyamada. M^3 Scaling Law: Optimizing Multi-Epoch, Multi-Lingual, and Multi-Stage Training for Low-Resource Language Models [arXiv]. arXiv 2026 (Under review)

7. Kosuke Kihara, Junki Mori, Taiki Miyagawa, Akinori F. Ebihara. クラス不均衡下連合型ソースフリードメイン適応における画像基盤モデルによる特徴量抽出器の再考 (Rethinking the Backbone in Class Imbalanced Federated Source Free Domain Adaptation: The Utility of Vision Foundation Models) [URL: TBA]. Symposium on Cryptography and Information Security (SCIS), 2026. (in Japanese)

6. Hideyuki Oiso, Kazuya Kakizaki, Taiki Miyagawa, Ryo Furukawa. LLM製知識グラフに対するEntity Alignmentの有効性評価 (Evaluation of Entity Aligners for LLM-generated Knowledge Graphs) [URL]. Information-Based Induction Sciences Workshop (IBIS), 2025. (in Japanese)

5. Akito Narahara, Taiki Miyagawa,  Akinori F. Ebihara. 早期時系列分類のためのH一致性をもつ損失関数 (H‑Consistent Loss Functions for Early Classification of Time Series) [URL]. Meeting on Image Recognition and Understanding (MIRU), 2025. (in Japanese)

4. Hiroshi Fukui, Taiki Miyagawa,  Akinori F. Ebihara. 追跡のためのCross-attentionモジュールを用いたクエリ伝搬アルゴリズムの軽量化 (Lightweight Query Propagation Algorithm Using a Cross‑Attention Module for Multi-Object Tracking) [URL]. Meeting on Image Recognition and Understanding (MIRU), 2025. (in Japanese)

3. Taiki Miyagawa,  Akinori F. Ebihara. 統計的早期時系列分類のレビュー：損失関数の性質 (Review: Loss Function for Statistical Early Classification of Time Series) [URL]. Meeting on Image Recognition and Understanding (MIRU), 2025. (in Japanese)

2. Batnyam Enkhtaivan, Satoru Momiyama, Taiki Miyagawa, Isamu Teranishi. コンバーターを用いた効率的な不均一モデル構造間の連合学習 (Efficient Federated Learning on Inhomogeneous Model Architectures Using Converters) [URL]. Forum on Data Engineering and Information Management (DEIM), 2025. (in Japanese)

1. Fumiya Morimoto, Taiki Miyagawa, Kazuya Kakizaki. 距離学習における敵対的訓練とデータ拡張との協調性：画像分類タスクの精度向上に向けて (Harmonization of Adversarial Training and Data Augmentation in Metric Learning: Toward Better Image Classification) [URL]. Meeting on Image Recognition and Understanding (MIRU), 2024. (in Japanese)

Patents (Selected)

Please search on WIPO PATENTSCOPE (Search query: ALLNAMES:(Taiki Miyagawa) AND PA:NEC* ).

21. Kazuhiro Watanabe, Yuji Iwadate, Masahiro Saikou, Akinori Ebihara, Taiki Miyagawa. IMAGE PROCESSING DEVICE, IMAGE PROCESSING METHOD, AND STORAGE MEDIUM. US20250078259.

20. Taiki Miyagawa. INFORMATION PROCESSING APPARATUS, INFORMATION PROCESSING METHOD, AND RECORDING MEDIUM. US20250021303.

19. Kapik Lee, Taiki Miyagawa, Akinori Ebihara. INFORMATION PROCESSING DEVICE, INFORMATION PROCESSING METHOD, AND RECORDING MEDIUM. WO2025009057.

18. Taiki Miyagawa, Akinori Ebihara. CLASSIFICATION APPARATUS, CLASSIFICATION METHOD, AND RECORDING MEDIUM. US20250005045.

17. Takaya Miyamoto, Akinori Ebihara, Taiki Miyagawa. INFORMATION PROCESSING DEVICE, INFORMATION PROCESSING METHOD, AND RECORDING MEDIUM. WO2024252527.

16. Akinori Ebihara, Taiki Miyagawa. INFORMATION PROCESSING DEVICE, INFORMATION PROCESSING METHOD, AND RECORDING MEDIUM. WO2024241383.

15. Taiki Miyagawa, Akinori Ebihara. INFORMATION PROCESSING DEVICE, INFORMATION PROCESSING METHOD, AND INFORMATION PROCESSING PROGRAM. WO2024189832.

14.  Taiki Miyagawa, Akinori Ebihara. INFORMATION PROCESSING DEVICE, INFORMATION PROCESSING METHOD, AND INFORMATION PROCESSING PROGRAM. WO2024189833.

13. Kazuhiro Watanabe, Akinori Ebihara, Taiki Miyagawa. INFORMATION PROCESSING APPARATUS, INFORMATION PROCESSING METHOD, MEDICAL IMAGE IDENTIFICATION DEVICE, AND NON-TRANSITORY COMPUTER READABLE MEDIUM STORING PROGRAM. US20240265536.

12. Akinori Ebihara, Taiki Miyagawa. INFORMATION PROCESSING DEVICE, INFORMATION PROCESSING METHOD, AND RECORDING MEDIUM. WO2024154230.

11. Kazuhiro Watanabe, Masahiro Saikou, Akinori Ebihara, Taiki Miyagawa. IMAGE PROCESSING DEVICE, IMAGE PROCESSING METHOD, AND STORAGE. US20240203086.

10. Hiroshi Fukui, Akinori Ebihara, Taiki Miyagawa. INFORMATION PROCESSING DEVICE, INFORMATION PROCESSING METHOD, AND RECORDING MEDIUM. WO2024111113.

9. Kazuhiro Watanabe, Yuji Iwadate, Masahiro Saikou, Akinori Ebihara, Taiki Miyagawa. Image Processing Device, Image Processing Method, and Storage Medium. US20240161294.

8. Kazuhiro Watanabe, Yuji Iwadate, Masahiro Saikou, Akinori Ebihara, Taiki Miyagawa. IMAGE PROCESSING DEVICE, IMAGE PROCESSING METHOD, AND STORAGE MEDIUM. US20240135539.

7. Kazuhiro Watanabe, Yuji Iwadate, Masahiro Saikou, Akinori Ebihara, Taiki Miyagawa. IMAGE PROCESSING DEVICE, IMAGE PROCESSING METHOD, AND STORAGE MEDIUM. US20240127443.

6. Kazuhiro Watanabe, Yuji Iwadate, Masahiro Saikou, Akinori Ebihara, Taiki Miyagawa. IMAGE PROCESSING DEVICE, IMAGE PROCESSING METHOD, AND STORAGE MEDIUM. US20240127442.

5. Akinori Ebihara, Taiki Miyagawa. INFORMATION PROCESSING DEVICE, INFORMATION PROCESSING METHOD, AND RECORDING MEDIUM. WO2024079853.

4. Akinori Ebihara, Taiki Miyagawa. INFORMATION PROCESSING DEVICE, INFORMATION PROCESSING METHOD, AND RECORDING MEDIUM. WO2024079854.

3. Kazuhiro Watanabe, Akinori Ebihara, Taiki Miyagawa. INFORMATION PROCESSING DEVICE, INFORMATION PROCESSING METHOD, MEDICAL IMAGE IDENTIFICATION DEVICE, AND NON-TRANSITORY COMPUTER-READABLE MEDIUM IN WHICH PROGRAM IS STORED. EP4354322.

2. Kazuhiro Watanabe, Yuji Iwadate, Masahiro Saikou, Akinori Ebihara, Taiki Miyagawa. IMAGE PROCESSING DEVICE, IMAGE PROCESSING METHOD, AND STORAGE MEDIUM. WO2024075411.

1. Akinori Ebihara, Taiki Miyagawa. LEARNING DEVICE, LEARNING METHOD, AND RECORDING MEDIUM. WO2021229663.

And more... Please search on WIPO PATENTSCOPE (Search query: ALLNAMES:(Taiki Miyagawa) AND PA:NEC* ).

Media

• April, 2025. NEC official press release on LinkedIn

Learning the Optimal Stopping for Early Classification within Finite Horizons via Sequential Probability Ratio Test

• January, 2025. NEC official press release on LinkedIn

Physics-informed Neural Networks for Functional Differential Equations: Cylindrical Approximation and Its Convergence Guarantees

• February 23, 2023. AIDB (web article)   

深層ニューラルネットワークの学習プロセスを運動方程式で表すことに成功　研究者本人が解説（NEC宮川大輝氏）

• May 14th, 2021. Nikkei xTech（日経クロステック追加取材）

  「ひらめき」と「慎重さ」が同居するNECの新AI、分析速度最大20倍の裏側

• May 10th, 2021. Dempa Publications (電波新聞)  

早押しクイズをヒントにAI開発　NECが顔認証やサイバー攻撃の検知に、処理時間20倍

• May 10th, 2021. Nikkei xTech  

NECが「早押しクイズ」型の顔認証AI、照合データの信頼性に応じ最大20倍速で判別

• May 7th, 2021. NEC official press release  

NEC、複雑な意思決定を行う際の脳活動の知見を応用したAI技術を開発 ～「早押しクイズ」にヒントを得て、顔認証やサイバー攻撃の分析を最大20倍高速化～

• May 7th, 2021. Nikkei (日本経済新聞, 朝刊, 10面, 530文字) 

NEC「一瞬」で顔認証、最短0.1秒未満、従来の20分の1、精度保ち、利便性向上

• May 7th, 2021. Cloud Watch (クラウドwatch) 

NEC、時系列データのリアルタイム分析向け新技術を開発　高い精度を維持しながら高速な判断を可能に

• May 7th, 2021. EnterpriseZine 

NEC、「早押しクイズ」をヒントにAI技術を開発

• May 7th, 2021. Robotstart (ロボスタ) 

簡単な問題は素早く回答、難問はじっくり考えてから答えを出すAIをNECが開発　顔認証の処理を最大20倍高速化　脳科学にヒント

• May 7th, 2021. ASCII×AI 

NEC、高い精度を維持しながら高速で判断するAI技術を開発

• May 7th, 2021. Yahoo News 

NEC、高い精度を維持しながら高速で判断するAI技術を開発

• May 6th, 2021. Nikkei Online (日経電子版) 

NEC、顔認証の判定時間短縮　最大で20分の1に

Intra Mentoring at NEC

• 2025    NEC Intern: Ryugo Morita, Hosei University, Japan. (1 patent)

• 2024    NEC Intern: Akito Narahara, Science Tokyo, Japan. (1 patent)

• 2024    NEC Intern: Yukai Xu, Kyushu University, Japan.

• 2023    NEC Intern: Kai Waseda, University of Tokyo, Japan. (1 patent)

• 2023    NEC Intern: Fumiya Morimoto, Kagoshima University, Japan. (1 domestic conference paper (MIRU 2024))

• 2023    NEC Intern: Daiki Nishiyama, Tokyo Institute of Technology (Science Tokyo), Japan.

• 2023    NEC Intern: Hideyuki Oiso, Tokyo Institute of Technology (Science Tokyo), Japan. (1 patent, 1 international conference paper (Interspeech 2024))

Professional Service

Reviewer

• NeurIPS (Top 10% Reviewers in 2023), ICLR, ICML, AISTATS, AAAI, CVPR, ECCV, ARR, PAKDD, ICANN, and more.

Session Chair

• The 39th Annual Conference of the Japanese Society for Artificial Intelligence (JSAI 2025).

• The 38th Annual Conference of the Japanese Society for Artificial Intelligence (JSAI 2024).

• The 37th Annual Conference of the Japanese Society for Artificial Intelligence (JSAI 2023).

Others

• Expo 2025 Osaka, Kansai: Our model, SPRT-TANDEM (ICLR'21, ICML'21, ICASSP'23, & ICLR'25), was exhibited at the Sumitomo pavilion, Expo 2025 Osaka, Kansai. [URL] [My tweet (JP)]

• MIRU 2023: A short talk in a company planning event ("NEC’s Cutting-Edge AI & Computer Vision Research and Its Future") at the 28th Meeting on Image Recognition and Understanding (MIRU 2023), Japan. [URL]