論文 - 西　宏章

pvFed: Personalized Vertical Federated learning for Client-Specific Tasks

Nishikawa A., Yanabe T., Sakuma Y., Okuda Y., Nishi H.

IEEJ Transactions on Electrical and Electronic Engineering （IEEJ Transactions on Electrical and Electronic Engineering）  19 （ 6 ） 993 - 1000 2024年06月

ISSN  19314973

　概要を見る

Federated Learning (FL) is a distributed machine learning paradigm that enables multiple data holders to collaborate on building machine learning models while preserving the privacy of their data. FL can be categorized as horizontal or vertical, depending on the distribution characteristics of the data. Specifically, horizontal FL uses data partitioned in the sample space, whereas vertical FL uses data partitioned in the feature space. Traditional vertical FL methods aim to facilitate collaboration among clients to infer a single global target. However, these methods may be impractical because each client often has a unique target to be inferred. In this paper, we propose a novel vertical FL method, called personalized vertical federated learning (pvFed), which addresses this limitation by allowing each client to perform inferences specific to their individual task. To the best of our knowledge, no existing method currently addresses this limitation. The objective of pvFed is to construct a global model that generates a representation vector to support client inference. The global model, constructed using distillation and dimensionality reduction, takes a sample ID common to all clients as input and outputs a sample-specific representation vector. Clients utilize the intermediate representation of their own model and the representation vectors output by the global model for inference. Because these vectors are not dependent on client-specific tasks, clients can repurpose them for any additional tasks. Our experiments, conducted on two distinct data types—image and tabular data sets, under a vertical partitioning where each client had its own specific task, demonstrated the efficacy of vectors generated by the global model in pvFed for client inference. © 2024 Institute of Electrical Engineer of Japan and Wiley Periodicals LLC.

• Access to Document (DOI)

FPGA Live Migration of Content-Based Anonymization Mechanism with Network Consistency

Toyohara T., Yamamoto K., Nishi H.

IEEJ Transactions on Electrical and Electronic Engineering （IEEJ Transactions on Electrical and Electronic Engineering）  19 （ 4 ） 507 - 517 2024年04月

ISSN  19314973

　概要を見る

In smart communities, data are aggregated, and various services are provided based on it. When considering the optimization of service delivery through cooperative processing between the edge and the cloud in smart communities, it is essential to realize network transparent edge services due to the constraints of not touching end devices in order to improve user convenience and additional cost reduction. Moreover, considering the minimization of service provision and service delivery latency at the edge, it is desirable to use field-programmable gate arrays (FPGAs), which have lower power consumption, lower latency, and higher throughput than conventional processors. Additionally, to advance towards even lower latency, the distribution of processing load is important, and service migration is necessary for this purpose. Service migration is generally used to increase service availability, and even for edge services, service migration is important because the distribution of processing and the location where the service should be provided changes over time due to user mobility. When considering network transparency, service migration should be performed on the communication path between the end device and the cloud. Furthermore, to enhance service transparency, it is essential to maintain communication consistency for continuous service provision and minimize service downtime. Therefore, this study proposes a migration mechanism with FPGAs to satisfy these requirements. As an example of a real application, we target a network-transparent data anonymization application to protect the privacy of data providers and evaluate the proposed architecture based on the reality of latency and other factors. © 2023 Institute of Electrical Engineer of Japan and Wiley Periodicals LLC.

• Access to Document (DOI)

Network-Transparent Service Container and Migration with Information Distribution Management in Edge Environments

Namba Y., Shimahara S., Nishi H.

ACM International Conference Proceeding Series    23 - 28 2024年01月

　概要を見る

Smart communities use myriad IoT devices and the data they transmit to perform global optimization. In network-transparent service provision, data transmitted from IoT devices are analyzed using network devices such as switches and routers in the edge region to add value to the data without modifying the IoT devices. However, to establish a mechanism for providing such network transparency services while complying with the opt-in or opt-out of data provided by users, data distribution management is necessary to determine the stored location and path of the target data. In addition, because a smart community environment is assumed to be a heterogeneous computing-resource environment, it is necessary to envision a dynamic relocation function for transparent network services considering the usage of computing resources. In this study, we proposed a method for managing and migrating transparent network services deployed on edge nodes using containers and linked it to an existing information distribution management mechanism that enables data distribution management at the service level. In the experiments, we confirmed that dynamic placement was possible under a load. We also measured the response delay associated with the provision of a transparent network service and confirmed that the associated processing delay was approximately 10 ms.

• Access to Document (DOI)

Preface

Li Y., Nishi H., Pang P.

ACM International Conference Proceeding Series 2024年01月

Anonymity‐Aware Framework for Designing Recommender Systems

M Honda, H Nishi

IEEJ Transactions on Electrical and Electronic Engineering 2024年

Anonymity-Aware Framework for Designing Recommender Systems

Honda M., Nishi H.

IEEJ Transactions on Electrical and Electronic Engineering （IEEJ Transactions on Electrical and Electronic Engineering）  19 （ 9 ） 1455 - 1464 2024年

ISSN  19314973

　概要を見る

Due to increasing secondary use of data, recommender systems using anonymized data are in demand. However, implementing a recommender system requires complicated data processing and programming, and the relationship between anonymization level and recommendation quality has not been investigated. Therefore, this study proposes a framework that facilitates the development of recommender systems. Additionally, a method is proposed for quantitatively evaluating recommendation quality when the anonymization level is varied. The proposed method promotes data utilization by recommender systems and determination of compensation for providing data based on anonymization level. © 2024 Institute of Electrical Engineer of Japan and Wiley Periodicals LLC.

• Access to Document (DOI)

FPGA Live Migration of Content‐Based Anonymization Mechanism with Network Consistency

T Toyohara, K Yamamoto, H Nishi

IEEJ Transactions on Electrical and Electronic Engineering 19 (4), 507-517 2024年

pvFed: Personalized Vertical Federated learning for Client‐Specific Tasks

A Nishikawa, T Yanabe, Y Sakuma, Y Okuda, H Nishi

IEEJ Transactions on Electrical and Electronic Engineering 2024年

Proceedings of the 2024 7th International Conference on Software Engineering and Information Management, ICSIM 2024, Suva, Fiji, January 23-25, 2024

ICSIM （ACM）  2024年

• Access to Document (DOI)

Proposal of Differential Privacy Anonymization for IoT Applications Using MQTT Broker

Morise K., Toyohara T., Nishi H.

Proceedings - IEEE Consumer Communications and Networking Conference, CCNC    634 - 635 2024年

ISSN  23319860

　概要を見る

IoT applications require secure communication methods that protect personal information contained in communication data. This study focuses on MQTT, a low-cost protocol used for IoT communication, and proposes a mechanism to anonymize communication data between IoT and clients. MQTT is a publish-subscribe model of communication where a broker handles many-to-many communications among clients. Due to the concentration of communications on the broker, it is efficient to anonymize data there. Therefore, the proposed mechanism performs differential privacy anonymization of communication data on the MQTT broker. We also propose a mechanism to anonymize data according to anonymization criteria required by senders and receivers using topic names and user properties, which are features of MQTT. We implemented the proposed mechanism in an FPGA-based MQTT broker and confirmed that it achieves the same throughput and low latency as regular MQTT communication and satisfies IoT applications such as power control and automated driving that require sub-millisecond latency.

• Access to Document (DOI)

Practical Trajectory Anonymization Method Using Latent Space Generalization

Sakuma Y., Nishi H.

IEEJ Transactions on Electrical and Electronic Engineering 2024年

ISSN  19314973

　概要を見る

The global positioning system (GPS) data are commonly used for location-based services such as traffic flow prediction. However, such data contain considerable sensitive information and thus, they must be anonymized before being published. In this study, we investigate trajectory anonymization. Previous methods have limitations in that they cannot be applied for different load network sparseness and cannot preserve the trajectory information. Thus, we propose a DNN-based method that can anonymize trajectories with different load network sparseness and also preserve the trajectory information. Specifically, the trajectories are projected to the latent space using the pre-trained encoder-decoder model, and the latent variables are generalized. Furthermore, to reduce the information loss, we propose a segment-aware trajectory modeling and study the effectiveness of assuming the normal distribution to the latent space. The experimental results using real GPS data show the effectiveness of the proposed method, presenting the improvement in the data reservation rate by approximately 3% and reducing the reconstruction error by approximately 31%. © 2024 The Author(s). IEEJ Transactions on Electrical and Electronic Engineering published by Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.

• Access to Document (DOI)

Network-Transparent Load Balancing and Migration for Edge Computing

Namba Y., Morishima R., Nishi H.

Proceedings - IEEE Consumer Communications and Networking Conference, CCNC    228 - 233 2024年

ISSN  23319860

　概要を見る

The network infrastructure in a smart community uses the cloud and compute servers, and the edge and fog areas to perform processing in the optimal place. Nodes in the edge and fog areas have various processing capacities, and there must be load-balancing mechanisms to meet the requirements of applications. However, Internet of Things (IoT) devices are generally small, computationally constrained, and numerous, so reconfiguration and functionality updates for the network must be kept to a minimum. In addition, if nodes establish connections with other nodes to exchange their information, the number of connections will be large, and communication costs will be high. In this study, we use nodes called 'Distributor' that can monitor data flow through the network without changing the existing infrastructure. Piggyback technique is used to add information onto the packets flowing in the network. Distributors add the information to the packets and exchange information to each other without establishing connections with other nodes, and they perform load balancing based on the information. By changing and migrating the processing locations among Distributors, IoT devices can continue to communicate with other nodes even during load balancing. Experiments confirmed that load balancing could be performed to prepare for sudden load increases and that the impact of piggybacking on occupied bandwidth and processing delay was 0.177% and 0.06%, respectively. We also confirmed that sensor data can be sent without reconnecting communications, even during load balancing.

• Access to Document (DOI)

Accelerator for Trajectory Anonymization Using Map Matching

Nakano H., Yamamoto K., Nishi H.

IEEE Access 12   152277 - 152288 2024年

ISSN  2169-3536

　概要を見る

This study is focused on hardware-based anonymization of location data for addressing privacy concerns while maintaining data utility in applications such as traffic-congestion prediction. The anonymization process involves approximating user location data to map intersections, identifying trajectories, and applying k-anonymization. However, the processing time in software implementation should be minimized. For real-time services, the process must be completed within 30 s to 1 min, which corresponds to the information update interval. Service requirements cannot be met when the volume of data and the scale of maps increase. To address this issue, we implemented a hardware accelerator on an field programmable gate array to perform the process with the aim of reducing the processing time. As a result, the maximum throughput of the proposed mechanism was approximately 150,000 data/s, which surpasses the target value of 130,000 data/s. The data retention rate, which represents information loss, was approximately 90% for the commonly used range of 10≤ k≤ 20. The difference in the data retention rate compared with the software implementation was approximately 1%. These results confirmed the effectiveness of the anonymization results obtained using the proposed mechanism.

• Access to Document (DOI)

A Low-Code Development Platform for Industrialized Insect-based Bioconversion Systems

Pang C., Liang J., Yu G., Zhang J., Nishi H., Wang K.I.K., Li Z., Dai W., Vyatkin V.

IEEE International Symposium on Industrial Electronics 2024年

ISSN  21635137

　概要を見る

Industrialized insect-based bioconversion is a technologically promising and economical viable approach for solving global crisis of food security and waste management. The successful implementation of such a multi-disciplinary system requires comprehensive expertise and skills from biotechnology, automation, information, and circular economy domains. This is rather difficult for existing practitioners. To simplify their jobs, by leveraging recent advances in industrial informatics, this work proposed and developed a low-code development platform for engineering the automation logic of industrial insect rearing and processing. For feasibility demonstration, a pilot facility has been designed, deployed, and benchmarked in this work. When compared with the conventional pro-code practice, the proposed low-code approach can significantly reduce development costs, shorten delivery times, and improve communication efficiency.

• Access to Document (DOI)

Access Control Management System for Edge Computing Environment Using Tag-Based Matching and Cache Injection

Shimahara S., Nohara M., Nishi H.

IEEJ Transactions on Electrical and Electronic Engineering （IEEJ Transactions on Electrical and Electronic Engineering）  18 （ 9 ） 1472 - 1481 2023年09月

ISSN  19314973

　概要を見る

With increasing attention to the secondary use of data, privacy protection laws such as the General Data Protection Regulation (GDPR) and California Consumer Privacy Act (CCPA) have come into force to protect residents from unauthorized exploitation of personal data. Flow management is an effective method to track the distribution of user data in a multi-domain network environment. Transparency in the relationship between data and service providers is essential to track the status of data provision. Vendor and Consumer Relationship Management (VCRM) manages this relationship by regulating relevant flow information. VCRM is expected to collaborate with the edge-computing environment in the management of regional unit flow, wherein all access to personal user data should be restricted in accordance with the stakeholders' requests. Accordingly, this study proposes a packet processing function that collaborates with VCRM to manage the access control of flows passing through a regional network. Moreover, this paper proposes methods to reduce packet processing time on the switch, and requests to the switch controller. Using tag-based matching and cache injection methods, the number of requests to the controller decreased to 33.4% of the original figure. Our proposed method also reduces the total processing time on flow management switches in the network environment in the analysis of encrypted TCP streams such as HTTP over TLS and MQTT over TLS. © 2023 Institute of Electrical Engineer of Japan and Wiley Periodicals LLC.

• Access to Document (DOI)

COVID-19 mRNA vaccination status and concerns among pregnant women in Japan: a multicenter questionnaire survey

Takahashi, K; Samura, O; Hasegawa, A; Okubo, H; Morimoto, K; Horiya, M; Okamoto, A; Ochiai, D; Tanaka, M; Sekiguchi, M; Miyasaka, N; Suzuki, Y; Tabata, T; Hayata, E; Nakata, M; Suzuki, T; Nishi, H; Toda, Y; Tanigaki, S; Furuya, N; Hasegawa, J; Tamaru, S; Kamei, Y; Sayama, S; Nagamatsu, T; Takahashi, YO; Kitagawa, M; Arakaki, T; Sekizawa, A

BMC PREGNANCY AND CHILDBIRTH 23 （ 1 ）  2023年05月

• Access to Document (DOI)

Anthropometric measurements of term singletons at 6 years of age born from fresh and frozen embryo transfer: A multicenter prospective study in Japan

Ueno, K, Kojima, J, Suzuki, K, Kuwahara, A, Higuchi, Y, Tanaka, A, Utsunomiya, T, Mio, Y, Nishi, H, Yoshimura, Y, Irahara, M, Kuji, N

REPRODUCTIVE MEDICINE AND BIOLOGY 22 （ 1 ）  2023年01月

ISSN  1445-5781

• Access to Document (DOI)

有用性保持を目的とした多次元データとしての経路情報匿名化

佐久間， 結子， 岩井， 智夢， 西川， 誠人， 西宏章

AIC カンファレンス予稿集, 55-55 2023年

マルチチャンネル音源を使用した異常音検知・分離・音源同定手法の提案と大型冷凍機を用いた評価

西川， 誠人， 服部， 一裕， 田中， 基雅， 村並， 広章， 西宏章

AIC カンファレンス予稿集, 59-59 2023年

スマートシティ情報インフラは FPGA を求めるか?-スマートシティをやるなら FPGA が必要といわせたい

西宏章

研究報告システムと LSI の設計技術 (SLDM) 2023 (5), 1-1 2023年