Search

Federated Learning, a model-sharing method, and Data Collaboration, a non-model-sharing method, are recognized as data analysis methods for distributed data. In Federated Learning, clients send only the parameters of a machine learning model to the central server. In Data Collaboration, clients send data that has undergone irreversibly transformed through dimensionality reduction to the central server. Both methods are designed with privacy concerns, but privacy is not guaranteed. Differential Privacy, a theoretical and quantitative privacy criterion, has been applied to Federated Learning to achieve rigorous privacy preservation. In this paper, we introduce a novel method using PCA (Principal Component Analysis) that finds low-rank approximation of a matrix preserving the variance, aiming to apply Differential Privacy to Data Collaboration. Experimental evaluation using the proposed method show that differentially-private Data Collaboration achieves comparable performance to differentially-private Federated Learning.

The research area of cybersecurity covers a wide range of fields from networking, software, and hardware to cryptography, authentication, and cyberattack countermeasures. Meanwhile, few cybersecurity experts are familiar with all areas of cybersecurity research. So, an evidence-based analysis covering all fields is very important to understand research trends in cybersecurity without bias. Such a field of study is called scientometrics, and there are many studies of scientometrics in the area of cybersecurity. However, in the analysis of these existing studies, the overall structure of the cybersecurity research is not clarified, and it is difficult to grasp the overall picture of cybersecurity research. In this study, we focus on cybersecurity as a research area covered by scientometrics analysis and conduct a detailed analysis of research trends in cybersecurity using e-CSTI (evidence data platform constructed by Council for Science, Technology and Innovation), and Dimensions bibliographic data for the 10 years from 2010 to 2019. Especially, we identify four representative research clusters (cyberattack, cryptography, authentication, and blockchain) and 55 research sub-clusters in the area of cybersecurity. We analyze the research trends in each country and the trends of the research topics of interest at the cluster level. For example, our results show that there are differences in research topics of emphasis between the U.S. and China. e-CSTI assists policymakers and researchers in getting a comprehensive understanding of global research trends and topics of interest in cybersecurity research.

In recent years, the convenience and potential use of crypto-assets such as Bitcoin and Ethereum have attracted increasing attention. On the other hand, there have been reports of attacks on the blockchain networks that support crypto-assets in an attempt to steal other users’ assets. In the past, research on attack observation against blockchains has used techniques such as holding real crypto-assets to lure attackers into honeypots or falsifying balances to attackers. However, these methods risk losing crypto-assets to attackers or being exposed as honeypots to attackers. To solve these problems, we propose a new RPC (Remote Procedure Call) honeypot method that returns the wallet address of another party holding a high balance in response to an attacker’s request, thereby luring the attacker without having the real crypto-assets. Our experimental evaluation shows that this method can attract more attackers than the method with zero-balance wallets and can observe more sophisticated attacks. Furthermore, we proposed a risk reduction strategy for crypto-asset theft by applying the idea of our method. In the log analysis process, we devised a new clustering method using the number of times an attacker executes a specific method as a feature. By applying this method, we successfully classified attackers based on their objectives, demonstrating the efficient analysis of vast amounts of log data.