Authorship：Lead author   Language：Japanese  

researchmap

Authorship：Lead author   Language：Japanese   Publishing type：Research paper (scientific journal)   Publisher：Information Processing Society of Japan  

This paper clarifies the usefulness and limitations of machine unlearning about data protection legislation. The machine unlearning efficiently removes specific data from AI pretrained models and plays a vital role in responding to data protection legislation. EU General Data Protection Regulation (GDPR) stip- ulates the right to be forgotten, but the legal system differs in other regions. Therefore, the data protection legislation of each jurisdiction (EU, United States, India, and Japan) is analyzed, and the legal applicability of machine unlearning are evaluated. As a result, more effective use of machine unlearning is expected in the development of the guidelines and legislation in the future.

This work is published under the license of the Information Processing Society of Japan (IPSJ), the copyright holder of this work. Use of this work is subject to the “Copyright Regulations of the Information Processing Society of Japan” and the Copyright Law.

File： マシン・アンラーニングの法的意義と限界_データ保護法制との関係性の検討.pdf

DOI： 10.20729/0002004348

researchmap

Authorship：Lead author   Language：Japanese   Publisher：Technical Information Association  

This article provides an explanation of the “Act on the Promotion of Research, Development, and Utilization of Artificial Intelligence Technologies” (the “AI Act”), which came into effect in Japan on June 4, 2025. While the Act places emphasis on promoting AI research, development, and utilization, it also establishes an institutional foundation for risk management. Its distinguishing feature lies in adopting a uniquely Japanese approach.

File： 日本版AI新法の静かなる施行_研究開発2508月号09.pdf

researchmap

Authorship：Lead author   Language：Japanese   Publishing type：Research paper (conference, symposium, etc.)   Publisher：Information Processing Society of Japan  

The EU AI Act's Prohibited AI Practices, which came into effect in February 2025, prohibits the use of AI systems for the purpose of inferring the emotions of natural persons in the workplace and in educational institutions. The specific details of the regulation are clarified in the Guidelines for Prohibited AI Practices. On the other hand, in Japan, there are no clear guidelines or regulations on emotion inference technology, and there have been some examples of demonstration experiments and introduction of such technology into products and services. In this paper, I examine how the protection of personal information should be handled in emotion inference technology. This paper clarifies that the regulations in the EU suggest the necessity of careful institutional design in Japan as well, taking into account the purpose of use, technical limitations, and the relationship with inner freedom.

This work is published under the license of the Information Processing Society of Japan (IPSJ), the copyright holder of this work. Use of this work is subject to the “Copyright Regulations of the Information Processing Society of Japan” and the Copyright Law.

File： EU AI法における感情推定規制と内面情報保護の在り方IPSJ-EIP25108001.pdf

researchmap

Authorship：Lead author   Language：Japanese  

The Japanese court decision on May 16, 2024 (the “DABUS Invention Case”), in which the applicability of AI as an “inventor” was disputed, provided an important opportunity to question how Japanese patent law should respond to the development of AI. The day is not far off when “AI fully autonomous inventions” will be realized, which will not require any human creative involvement. The “inventor” as defined in patent law is interpreted to be limited to a “natural person,” but with the future evolution of AI, how to treat “AI fully autonomous inventions” will be a major issue. The legal issues concerning “AI fully autonomous inventions” are discussed by comparing the decision of the German Federal Court of Justice in the “DABUS Invention Case” on June 11, 2024 with court decisions in Japan. Under the current Japanese patent law, “AI fully autonomous inventions” may cause the problem of absence of “inventor” or presumption. Therefore, we believe that there is a limit to the solution of this problem by the theory of interpretation, and that it is necessary to consider the problem by the legislation.

researchmap

Authorship：Lead author   Language：Japanese  

File： 24.12.20研究開発2412月号07_鈴木健二R.pdf

researchmap

Language：English  

The use of large-scale, web-scraped datasets to train face recognition models has raised significant privacy and bias concerns. Synthetic methods mitigate these concerns and provide scalable and controllable face generation to enable fair and accurate face recognition. However, existing synthetic datasets display limited intraclass and interclass diversity and do not match the face recognition performance obtained using real datasets. Here, we propose VariFace, a two-stage diffusion-based pipeline to create fair and diverse synthetic face datasets to train face recognition models. Specifically, we introduce three methods: Face Recognition Consistency to refine demographic labels, Face Vendi Score Guidance to improve interclass diversity, and Divergence Score Conditioning to balance the identity preservation-intraclass diversity trade-off. When constrained to the same dataset size, VariFace considerably outperforms previous synthetic datasets (0.9200 → 0.9405) and achieves comparable performance to face recognition models trained with real data (Real Gap = -0.0065). In an unconstrained setting, VariFace not only consistently achieves better performance compared to previous synthetic methods across dataset sizes but also, for the first time, outperforms the real dataset (CASIA-WebFace) across six evaluation datasets. This sets a new state-of-the-art performance with an average face verification accuracy of 0.9567 (Real Gap = +0.0097) across LFW, CFP-FP, CPLFW, AgeDB, and CALFW datasets and 0.9366 (Real Gap = +0.0380) on the RFW dataset.

File： 2412.06235v1.pdf

DOI： 10.48550/arXiv.2412.06235

researchmap

Authorship：Lead author   Language：Japanese   Publishing type：Research paper (conference, symposium, etc.)   Publisher：Information Processing Society of Japan  

This paper discusses the importance of “legitimate interests” as a legal basis for data processing under the EU General Data Protection Regulation (GDPR) and the legal issues involved. Data processing based on legitimate interests is used in many situations, but its application is difficult to interpret legally because it is a general provision. However, legitimate interests require an understanding of the position of legitimate interests, including the impact of recent technological evolution, legislation in the EU, and changes in social acceptance. The purpose of this paper is to summarize the complex legal issues related to legitimate interests from the perspective of the use of AI learning data, based on recent EU legislative trends, court cases, academic theories, and the views of the data protection authorities in Italy, the Netherlands, and France. Although the three-step criteria of legitimacy, necessity, and proportionality have been used in recent cases, the specific legal interpretations are still unclear.

File： EU一般データ保護規則における「正当な利益」の法的課題 .pdf

researchmap

Authorship：Lead author   Language：Japanese   Publishing type：Research paper (conference, symposium, etc.)   Publisher：Information Processing Society of Japan  

Training data and learned models may contain personal information that should be erased based on legal requirements. However, erasing such data from the learned model and rebuilding the model is economically expensive. For this reason, machine unlearning has been attracting attention as a technique for forgetting specific personal information from learned models. This paper focuses on the significance and technical limitations of machine unlearning and clarifies its legal status. The EU's GDPR and the U.S. state of California's CCPA are open to flexible interpretations of the technical limitations.

This work is published under the permission of the Information Processing Society of Japan (IPSJ), the copyright holder. Use of this work is subject to the “Copyright Rules of the Information Processing Society of Japan” and copyright laws.

File： プライバシー保護におけるマシンアンラーニングの法的役割.pdf

researchmap

Authorship：Lead author   Language：Japanese   Publishing type：Research paper (conference, symposium, etc.)  

Large-scale language models are a high-performance technology that can bring about change, but their behavior is very complex. Also, there are potential challenges such as misinformation, bias, and subliminal manipulation. Explainable AI is a technology that makes black box models understandable to humans. In December 2023, the European AI Act incorporates regulations on infrastructure models, generative AI, and general-purpose AI. We will discuss what kind of legal role explainable AI has for large-scale language models.

File： 大規模言語モデルにおける説明可能なAIの法的役割.pdf

researchmap

Authorship：Lead author   Language：Japanese   Publishing type：Research paper (conference, symposium, etc.)  

Advances in generative AI, such as large language models and image generation AI, are bringing about revolutionary technological changes. On the other hand, there are concerns about ethical, legal, and social issues posed by generative AI, and it is necessary to consider the nature of co-creation between generative AI and humans in the information society.
This paper considers the issues based on the EU AI Act, a comprehensive regulatory act on AI. With the rapid development of generative AI, the European Parliament adopted the EU AI Act in June 2023, adding provisions on foundation models and generative AI to the amendment. In relation to the new development of generative AI and its relationship with society, this paper outlines the legal system including technical aspects regarding the obligations of providers of foundation models and generative AI. The paper examines how to make the best use of generative AI and create fair society in our country.

Copyright 2023. Kenji SUZUKI. All Rights Reserved.

File： 欧州AI法案から考える生成系AI.pdf

researchmap

Language：English  

Our goal is to develop fine-grained real-image editing methods suitable for real-world applications. In this paper, we first summarize four requirements for these methods and propose a novel diffusion-based image editing framework with pixel-wise guidance that satisfies these requirements. Specifically, we train pixel-classifiers with a few annotated data and then infer the segmentation map of a target image. Users then manipulate the map to instruct how the image will be edited. We utilize a pre-trained diffusion model to generate edited images aligned with the user's intention with pixel-wise guidance. The effective combination of proposed guidance and other techniques enables highly controllable editing with preserving the outside of the edited area, which results in meeting our requirements. The experimental results demonstrate that our proposal outperforms the GAN-based method for editing quality and speed.

File： 2212.02024v3.pdf

researchmap

Authorship：Lead author   Language：Japanese   Publishing type：Research paper (conference, symposium, etc.)  

I propose an explainable fairness method that can not only mitigate data bias but also make humans understand the reason. Machine learning algorithms have high-risk use cases, such as hiring and loan decision, that demand fairness, accountability, and transparency. Differences in AI predictions due to sensitive attributes, such as gender, race, and age, have become fairness issue. Although various methods of bias mitigation in AI have been proposed, there is a problem that conventional methods do not allow humans to intuitively understand on how basis data bias mitigation is performed. Therefore, I propose a method of bias mitigation by using explainable AI. My proposed method allows humans to understand how bias mitigation is achieved. Experimental results of applying this method to credit scoring by German Credit dataset shows that the statistical parity difference improved from -0.108 to -0.004 on gender.

File： JSAI2023_3Xin451.pdf

DOI： 10.11517/pjsai.JSAI2023.0_3Xin451

researchmap

Language：English  

Generative models, particularly GANs, have been utilized for image editing. Although GAN-based methods perform well on generating reasonable contents aligned with the user's intentions, they struggle to strictly preserve the contents outside the editing region. To address this issue, we use diffusion models instead of GANs and propose a novel image-editing method, based on pixel-wise guidance. Specifically, we first train pixel-classifiers with few annotated data and then estimate the semantic segmentation map of a target image. Users then manipulate the map to instruct how the image is to be edited. The diffusion model generates an edited image via guidance by pixel-wise classifiers, such that the resultant image aligns with the manipulated map. As the guidance is conducted pixel-wise, the proposed method can create reasonable contents in the editing region while preserving the contents outside this region. The experimental results validate the advantages of the proposed method both quantitatively and qualitatively.

File： 2212.02024v3.pdf

DOI： 10.48550/arXiv.2212.02024

researchmap

Authorship：Lead author   Language：Japanese  

researchmap

Authorship：Lead author   Language：Japanese  

File： 説明可能なAIを身近にするためのディープラーニングツール.pdf

DOI： 10.20729/00218782

researchmap

Authorship：Lead author   Language：Japanese   Publishing type：Research paper (conference, symposium, etc.)  

researchmap

Authorship：Lead author   Language：Japanese   Publishing type：Research paper (conference, symposium, etc.)  

researchmap

Authorship：Lead author   Publishing type：Research paper (conference, symposium, etc.)  

Identifying the influence of training data for data cleansing can improve the accuracy of deep learning. An approach with stochastic gradient descent (SGD) called SGD-influence to calculate the influence scores was proposed, but, the calculation costs are expensive. It is necessary to temporally store the parameters of the model during training phase for inference phase to calculate influence sores. In close connection with the previous method, we propose a method to reduce cache files to store the parameters in training phase for calculating inference score. We only adopt the final parameters in last epoch for influence functions calculation. In our experiments on classification, the cache size of training using MNIST dataset with our approach is 1.236 MB. On the other hand, the previous method used cache size of 1.932 GB in last epoch. It means that cache size has been reduced to 1/1,563. We also observed the accuracy improvement by data cleansing with removal of negatively influential data using our approach as well as the previous method. Moreover, our simple and general proposed method to calculate influence scores is available on our auto ML tool without programing, Neural Network Console. The source code is also available.

researchmap

Language：English  

While there exist a plethora of deep learning tools and frameworks, the fast-growing complexity of the field brings new demands and challenges, such as more flexible network design, speedy computation on distributed setting, and compatibility between different tools. In this paper, we introduce Neural Network Libraries (this https URL), a deep learning framework designed from engineer's perspective, with emphasis on usability and compatibility as its core design principles. We elaborate on each of our design principles and its merits, and validate our attempts via experiments.

File： 2102.06725v2.pdf

DOI： 10.48550/arXiv.2102.06725

researchmap

Authorship：Lead author   Language：English  

Identifying the influence of training data for data cleansing can improve the accuracy of deep learning. An approach with stochastic gradient descent (SGD) called SGD-influence to calculate the influence scores was proposed, but, the calculation costs are expensive. It is necessary to temporally store the parameters of the model during training phase for inference phase to calculate influence sores. In close connection with the previous method, we propose a method to reduce cache files to store the parameters in training phase for calculating inference score. We only adopt the final parameters in last epoch for influence functions calculation. In our experiments on classification, the cache size of training using MNIST dataset with our approach is 1.236 MB. On the other hand, the previous method used cache size of 1.932 GB in last epoch. It means that cache size has been reduced to 1/1,563. We also observed the accuracy improvement by data cleansing with removal of negatively influential data using our approach as well as the previous method. Moreover, our simple and general proposed method to calculate influence scores is available on our auto ML tool without programing, Neural Network Console. The source code is also available.

File： 2103.11807v2.pdf

DOI： 10.48550/arXiv.2103.11807

researchmap

Authorship：Lead author   Publishing type：Research paper (scientific journal)   Publisher：Wiley  

DOI： 10.1002/1521-3951(200103)224:1<139::aid-pssb139>3.0.co;2-o

researchmap

Publishing type：Research paper (scientific journal)   Publisher：Elsevier BV  

DOI： 10.1016/s1386-9477(99)00348-3

researchmap

Authorship：Lead author   Language：English   Publishing type：Research paper (scientific journal)   Publisher：American Institute of Physics  

We report structural and optical properties of GaSb/GaAs self-assembled quantum dots (QDs) grown by molecular beam epitaxy. The QDs, with nanometer-scale dimensions, were characterized by atomic force microscopy. Furthermore, in photoluminescence (PL) measurements the feature from the QDs was observed at ∼1.1 eV, clearly separated from that of the wetting layer at ∼1.3 eV. With increasing excitation power, the peak from the QDs displayed a large shift towards higher energy. In addition, the temperature dependence of PL yielded a large thermal activation energy, 130 meV, confirming the strong localization of excitons in the QDs.

researchmap

Publishing type：Research paper (scientific journal)   Publisher：Elsevier BV  

DOI： 10.1016/s0167-9317(99)00164-1

researchmap

Authorship：Lead author   Publishing type：Research paper (scientific journal)   Publisher：Elsevier BV  

DOI： 10.1016/s0022-0248(99)00021-4

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：OSA  

Injecting carriers by using scanning tunneling microscope (STM) and detecting the luminescence, we can achieve optical measurements with very high spatial resolution. We refer to this technique as scanning tunneling luminescence (STL).

DOI： 10.1364/qo.1999.qmd3

researchmap

Authorship：Lead author   Language：English   Publishing type：Research paper (international conference proceedings)  

researchmap

Authorship：Lead author   Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Optical Society of America (OSA)  

DOI： 10.1364/qo.1999.qmc2

researchmap

Authorship：Lead author   Language：English   Publishing type：Doctoral thesis   Publisher：University of Tokyo  

File： 鈴木健二博士論文.pdf

DOI： 10.11501/3162766

researchmap

Other Link： https://repository.dl.itc.u-tokyo.ac.jp/records/8819

Publishing type：Research paper (scientific journal)   Publisher：American Physical Society (APS)  

DOI： 10.1103/physrevb.58.r10147

researchmap

Other Link： http://harvest.aps.org/v2/journals/articles/10.1103/PhysRevB.58.R10147/fulltext

Authorship：Lead author  

researchmap

Authorship：Lead author   Language：English   Publisher：CRC Press  

DOI： 10.1201/9781003063100

researchmap

Publishing type：Research paper (scientific journal)   Publisher：AIP Publishing  

DOI： 10.1063/1.122174

researchmap

Authorship：Lead author   Language：English  

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

researchmap

Publishing type：Research paper (scientific journal)   Publisher：AIP Publishing  

DOI： 10.1063/1.121919

researchmap

Publishing type：Research paper (scientific journal)   Publisher：Elsevier BV  

DOI： 10.1016/s0038-1101(97)00305-5

researchmap

Publishing type：Research paper (scientific journal)   Publisher：Elsevier BV  

DOI： 10.1016/s0038-1101(97)00306-7

researchmap

Publishing type：Research paper (scientific journal)   Publisher：AIP Publishing  

DOI： 10.1063/1.121480

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

researchmap

Authorship：Lead author   Publishing type：Research paper (scientific journal)   Publisher：IOP Publishing  

DOI： 10.1143/jjap.37.l203

researchmap

Authorship：Lead author   Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/cleopr.1997.610550

researchmap

Language：English  

DOI： 10.1016/0022-0248(95)80172-9

researchmap

Total pages：389p   Responsible for pages：257-261   Language：Japanese  

CiNii Books

researchmap

Total pages：94   Responsible for pages：51-52  

ASIN

researchmap

Authorship：Lead author   Language：Japanese   Publishing type：Meeting report   Publisher：情報処理学会  

2024 年 3 月8 日（金），東京工業大学データサイエンス・ AI 全学教育機構主催の公開シンポジウム2024「DS・ AI を社会的側面から考える～社会のリーダーとなる人材と は」が開催された．東京工業大学大岡山キャンパス内の学 生支援・交流施設「Taki Plaza」に産官学および 学生を含む多方面からの参加者が集まった．参加者数は現地で 136 名（学士課程1 年生から博士後期課程までの 学生 63 名 含 む ）， オ ン ラ イ ン で 79 名，合計215 名となり， 大盛況であった．本シンポジウムは，昨年に続き 2 回目の開催である．昨年3 月に開催された第 1 回目のシ ンポジウムは，当機構の開設を記念する趣旨で行われた． 当大学は，全国で先駆けて2019 年からデータサイエンス・ AI 大学院全学教育に取り組んでおり，当機構は全学的な 教育組織として2022 年12 月に発足した．今回のシンポジ ウムは，「社会のリーダーとなる人材とは」というテーマで 行われた．将来，社会で活躍するためには，技術だけでな く社会の問題にも精通することが期待されている．新たな 取り組みとして，本 シ ン ポ ジ ウ ム で は 学 生 自 身 が「 AIと社会」 について考えることを主な目的とした．当機構の小野功副 機構長が司会を務め，当大学の益一哉学長の開会挨拶で 始まったシンポジウムは，2 時間40 分にわたり滞りなく進 行した．

本著作物は、著作権者である情報処理学会の利用許諾に基づき掲載しております。ご利用にあたっては、「情報処理学会著作権規定」および著作権法に従うものとします。

File： 会議レポート「DSAIを社会的側面から考える～社会のリーダーになる人材とは」.pdf

researchmap

Language：Japanese   Publisher：Sony  

researchmap

Language：Japanese   Presentation type：Public lecture, seminar, tutorial, course, or other speech  

File： 20211213_ATTA2021_Explainable_AI_Kenji_Suzuki.pdf

researchmap

Language：Japanese   Presentation type：Public lecture, seminar, tutorial, course, or other speech  

researchmap

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Symposium, workshop panel (nominated)  

Venue：Tokyo Institue of Technology, Taki Plaza   Country：Japan  

We practice education aimed at fostering a broad perspective that transcends the boundaries of the humanities and sciences. We introduce our graduate courses focusing on AI ethics, AI regulation, and its technology in the era of generative AI and mention its challenges.

Please contact us for materials: suzuki.k.ep@m.titech.ac.jp

File： AIと社会についての教育実践V6_distribute.pdf

researchmap

Language：Japanese   Presentation type：Poster presentation  

Country：Japan  

researchmap

Language：Japanese   Presentation type：Public lecture, seminar, tutorial, course, or other speech  

File： 生成系AIの社会課題V7_distribute.pdf

researchmap

Other Link： https://www.dsai.titech.ac.jp/news/news-1580/

Language：Japanese  

In the development of human resources through joint education between universities and companies, a complementary relationship is essential. Until now, educational programs have been designed with a focus on the complementary relationship between the theoretical and practical aspects of data science and AI. However, moving forward, it will be necessary to also emphasize the cultivation of a global perspective, economic value, and ethical values. In the EU, regulatory measures, including responses to ethical, legal, and social issues arising from generative AI, have been introduced under the "EU AI Act." Against this backdrop, I will explore the ideal form of joint education through industry-academia collaboration from the perspective of practitioner educators who are AI developers, legal scholars, and educators.

File： 生成AIへの法規制から考えるデータサイエンス・AIでの産学連携共同教育_鈴木健二V11.pdf

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

This article focuses on a case in which copyright management of musical compositions in a music school was contested. X and others who run a music school sued Y, a copyright management company, for a declaration that Y had no right to claim damages based on torts for copyright (performance rights) infringement against X and others for musical compositions managed by teachers and students in lessons. The first instance ruling confirmed the performance rights of musical compositions for the students' performances, but the appeal court denied it. The Supreme Court upheld the appeal court ruling and held that the performance rights do not extend to musical compositions performed by students in music schools, and that copyright management must be limited to teachers' performances. The standard of judgment did not use the "karaoke doctrine" that had been used in previous cases, but instead presented a general theory based on the "overall balance doctrine." This ruling indicated that when determining who is using musical compositions based on the form of performance, various circumstances such as the purpose and manner of the performance, and the content and degree of involvement in the performance should be taken into consideration.

File： 2024年度第7回 民事法研究会報告資料(最判令和4年10月24日民集76巻6号1348頁)鈴木健二R.pdf

researchmap

Language：Japanese   Presentation type：Symposium, workshop panel (nominated)  

Venue：Koube   Country：Japan  

AI training data and AI pretrained models may contain personal data that should be erased based on legal requirements. However, erasing pertinent data from AI pretrained models and rebuilding AI models involves economic costs. For this reason, machine unlearning has attracted attention as a technique to forget specific personal data from AI trained models. This presentation will focus on the significance and technical limitations of machine unlearning and clarify its legal status. The application of machine unlearning to privacy protection legislation in various countries and regions will also be discussed, and its usefulness and limitations will be noted.

Please cite the following as references when citing in your papers.
Kenji Suzuki, “Legal Role of Machine Unlearning in Privacy Protection,” Research Report Electronic Intellectual Property and Social Infrastructure (EIP), 2024-EIP-104, Vol. 17, pp. 1-6 (2024).
Download here.
https://researchmap.jp/kenjisuzuki/published_papers/46126935

File： マシン・アンラーニング技術の法的側面_鈴木健二V6.pdf

researchmap

Other Link： https://researchmap.jp/kenjisuzuki/published_papers/46126935

Language：Japanese   Presentation type：Symposium, workshop panel (nominated)  

Venue：Sapporo, Online   Country：Japan  

This presentation will explain how “joint education through industry-academia collaboration,” in which universities and companies cooperate with each other, can contribute to sustainable human resource development. Specifically, the lecture will address the role of practitioner faculty members, the importance of human resource development through industry-academia collaboration, issues in data science and AI education, career development and reskilling, and education in DS&AI interdisciplinary fields. Through this lecture, participants will share their understanding of the importance of developing human resources required by society and sustainable career development, as well as the state of education to acquire the skills necessary to play an active role in the digital society.

File： 東京科学大学と企業を結ぶ産学連携による共同教育：サステイナブルな人材育成（後編）_鈴木健二.pdf

researchmap

researchmap

Language：Japanese  

The EU AI Act, which was passed by the EU on May 21, 2024 and entered into force on August 1, 2024, is a comprehensive regulation on AI. The EU AI Act will gradually come into effect on February 2, 2025 for “prohibited AI” and on August 2, 2025 for “general-purpose AI models. While previously relying on voluntary regulations such as AI ethical guidelines, the “EU AI Act” has transformed them into legally binding regulations. Its scope of application covers a broad value chain, from the developers of AI models to their providers and operators.

In addition to the regulation based on the “risk-based approach” presented in the draft three years ago, a new regulation on General Purpose AI Models (GPAI) has also been introduced. Furthermore, it is important to note that the regulations are not limited to the EU, but also include extraterritorial application provisions, which apply to businesses outside the EU for products and services provided within the EU.

The EU AI Act is also attracting attention as a “Brussels effect,” in which EU regulations spill over into legal systems around the world. Deepening our understanding of it will have important implications for future research and development. In particular, for those involved in data science and AI R&D, this seminar will provide detailed and easy-to-understand explanations from my perspective as an AI R&D and information law researcher. We sincerely look forward to your participation.

File： 25.1.24_データサイエンティストのための「EU_AI法」_鈴木健二.pdf

researchmap

Other Link： https://us02web.zoom.us/webinar/register/WN_Dwbai6kGSj6HcCLmbv-PqA#/registration

Language：Japanese   Presentation type：Public lecture, seminar, tutorial, course, or other speech  

researchmap

Language：Japanese   Presentation type：Public lecture, seminar, tutorial, course, or other speech  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

Language：Japanese   Presentation type：Symposium, workshop panel (nominated)  

On March 12, 2026, I will deliver a lecture titled “AI Risks: Cybersecurity under the EU AI Act” at the Graduate University for Information Security's Special SSL Symposium for Corporate Legal Affairs. The EU has undergone a significant shift from AI ethics and voluntary guidelines to legally binding regulations with penalties, establishing an AI governance framework centered on the EU AI Act. This presentation will explain the cybersecurity requirements and accountability structures demanded of AI systems within this new regulatory framework. Specifically, high-risk AI systems require: risk management to assess potential dangers beforehand; preparation of technical documentation detailing the system's design and architecture; logging to track operational status; ensuring appropriate human oversight; and safe design incorporating accuracy, robustness, and cybersecurity. We will also overview the relationship between cyberattacks on AI (such as data contamination and adversarial inputs) and relevant EU legislation (GDPR, NIS2 Directive, Cyber Resilience Act). Compliance with the EU AI Act means not merely meeting regulations, but building corporate governance that enables the safe design and operation of AI, along with the ability to explain its safety. We hope this presentation offers practical insights for leveraging AI.

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

Language：Japanese   Presentation type：Public lecture, seminar, tutorial, course, or other speech  

researchmap

Language：Japanese  

researchmap

Language：English   Presentation type：Poster presentation  

researchmap

Language：English  

researchmap

Language：English  

researchmap

Language：Japanese   Presentation type：Public lecture, seminar, tutorial, course, or other speech  

researchmap

Language：English   Presentation type：Public lecture, seminar, tutorial, course, or other speech  

researchmap

Application no：PCT/JP2021/023319  Date applied：2021.6

Announcement no：WO/2022/009652  Date announced：2022.1

researchmap

Application no：PCT/JP2021/022041  Date applied：2021.6

Announcement no：WO/2022/024559  Date announced：2022.2

researchmap

Applicant：ソニーグループ株式会社

Application no：特願2021-056228  Date applied：2021.3

Announcement no：特開2022-153142  Date announced：2022.10

J-GLOBAL

researchmap

Application no：PCT/JP2021/012368  Date applied：2021.3

Announcement no：WO/2021/200503  Date announced：2021.7

researchmap

Applicant：ソニーグループ株式会社

Application no：特願2021-047996  Date applied：2021.3

Announcement no：特開2022-146822  Date announced：2022.10

J-GLOBAL

researchmap

Application no：PCT/JP2021/044541  Date applied：2021.3

Announcement no：WO/2022/163130  Date announced：2022.4

researchmap

Applicant：ソニーグループ株式会社

Application no：特願2021-024406  Date applied：2021.2

Announcement no：特開2022-126373  Date announced：2022.8

J-GLOBAL

researchmap

Application no：特願2021-555113(P2021-555113)  Date applied：2020.11

Patent/Registration no：特許7593328  Date registered：2024.11  Date issued：2024.12

Rights holder：Sony Group Corporation

researchmap

Application no：特願2005-84335(P2005-84335A)  Date applied：2005.3

Announcement no：特開2005-84335(P2005-84335A)  Date announced：2005.3

researchmap

Application no：特願2004-199518(P2004-199518)  Date applied：2004.7

Announcement no：特開2005-99731(P2005-99731A)  Date announced：2005.4

researchmap

Application no：特願2009-284653(P2009-284653)  Date applied：2004.3

Announcement no：特開2010-61171(P2010-61171A)  Date announced：2010.3

researchmap

Application no：特願2003-428873(P2003-428873)  Date applied：2003.12

Announcement no：特開2005-189385(P2005-189385A)  Date announced：2005.7

researchmap

Application no：特願2003-416599(P2003-416599)  Date applied：2003.12

Announcement no：特開2005-173469(P2005-173469A)  Date announced：2005.6

researchmap

Application no：特願2003-327744(P2003-327744)  Date applied：2003.9

Announcement no：特開2005-92013(P2005-92013A)  Date announced：2005.4

researchmap

Application no：特願2003-314162(P2003-314162)  Date applied：2003.9

Announcement no：特開2005-84227(P2005-84227A)  Date announced：2005.3

researchmap

Application no：特願2003-311098(P2003-311098)  Date applied：2003.9

Announcement no：特開2005-77971(P2005-77971A)  Date announced：2005.3

researchmap

Application no：特願2003-311672(P2003-311672)  Date applied：2003.9

Announcement no：特開2005-78022(P2005-78022A)  Date announced：2005.3

researchmap

Application no：特願2003-304822(P2003-304822)  Date applied：2003.8

Announcement no：特開2004-199032(P2004-199032A)  Date announced：2004.7

researchmap

Application no：特願2003-302261(P2003-302261)  Date applied：2003.8

Announcement no：特開2005-70573(P2005-70573A)  Date announced：2005.3

researchmap

Application no：特願2003-297849(P2003-297849)  Date applied：2003.8

Announcement no：特開2005-70235(P2005-70235A)  Date announced：2005.3

researchmap

Application no：特願2003-291999(P2003-291999)  Date applied：2003.8

Announcement no：特開2005-62444(P2005-62444A)  Date announced：2005.3

researchmap

Application no：特願2003-291405(P2003-291405)  Date applied：2003.8

Announcement no：特開特開2005-62394(P2005-62394A)  Date announced：2005.3

researchmap

Application no：特願2003-289693(P2003-289693)  Date applied：2003.8

Announcement no：特開2005-62297(P2005-62297A)  Date announced：2005.3

researchmap

Application no：特願2003-289694(P2003-289694)  Date applied：2003.8

Announcement no：特開2005-62298(P2005-62298A) 

researchmap

Application no：特願2003-288883(P2003-288883)  Date applied：2003.8

Announcement no：特開2005-55809(P2005-55809A) 

researchmap

Application no：特願2003-275485(P2003-275485)  Date applied：2003.7

Announcement no：特開2004-302401(P2004-302401A)  Date announced：2004.10

researchmap

Application no：特願2003-188923(P2003-188923)  Date applied：2003.6

Announcement no：特開2005-24798(P2005-24798A)  Date announced：2005.1

researchmap

Application no：特願2003-133307(P2003-133307)  Date applied：2003.5

Announcement no：特開2004-334118(P2004-334118A)  Date announced：2004.11

researchmap

Application no：特願2003-127818(P2003-127818)  Date applied：2003.5

Announcement no：特開2004-333728(P2004-333728A)  Date announced：2004.11

researchmap

Application no：特願2003-69185(P2003-69185)  Date applied：2003.3

Announcement no：特開2004-279620(P2004-279620A)  Date announced：2004.10

researchmap

Application no：特願2003-69187(P2003-69187)  Date applied：2003.3

Announcement no：特開2004-279621(P2004-279621A) 

researchmap

Application no：特願2003-69189(P2003-69189)  Date applied：2003.3

Announcement no：特開2004-279623(P2004-279623A)  Date announced：2004.10

researchmap

Application no：特願2003-69188(P2003-69188)  Date applied：2003.3

Announcement no：特開2004-279622(P2004-279622A)  Date announced：2004.10

researchmap

Application no：特願2004-264339(P2004-264339A)  Date applied：2003.1

Announcement no：特開2004-264339(P2004-264339A)  Date announced：2004.9

researchmap

Application no：特願2002-346176(P2002-346176)  Date applied：2002.11

Announcement no：特開2004-177816(P2004-177816A)  Date announced：2004.6

researchmap

Application no：特願2002-345001(P2002-345001)  Date applied：2002.11

Announcement no：特開2004-177730(P2004-177730A)  Date announced：2004.6

researchmap

Application no：特願2002-338781(P2002-338781)  Date applied：2002.11

Announcement no：特開2004-170830(P2004-170830A)  Date announced：2004.6

researchmap

Application no：特願2002-336168(P2002-336168)  Date applied：2002.11

Announcement no：特開2004-170668(P2004-170668A)  Date announced：2004.6

researchmap

Award type：Award from Japanese society, conference, symposium, etc.  Country：Japan

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

Type：Academic society, research group, etc. 

researchmap

Type：Competition, symposium, etc. 

researchmap

Type：Competition, symposium, etc. 

researchmap

Type：Peer review 

researchmap