Takamasa Isohara’s research while affiliated with KDDI Research and other places

Recently, an increasing number of laws have governed the useability of users' privacy. For example, Article 17 of the General Data Protection Regulation (GDPR), the right to be forgotten, requires machine learning applications to remove a portion of data from a dataset and retrain it if the user makes such a request. Furthermore, from the security perspective, training data for machine learning models, i.e., data that may contain user privacy, should be effectively protected, including appropriate erasure. Therefore, researchers propose various privacy-preserving methods to deal with such issues as machine unlearning. This paper provides an in-depth review of the security and privacy concerns in machine learning models. First, we present how machine learning can use users' private data in daily life and the role that the GDPR plays in this problem. Then, we introduce the concept of machine unlearning by describing the security threats in machine learning models and how to protect users' privacy from being violated using machine learning platforms. As the core content of the paper, we introduce and analyze current machine unlearning approaches and several representative research results and discuss them in the context of the data lineage. Furthermore, we also discuss the future research challenges in this field.

To improve the rate of taking security action, it is important to promote personalized approaches for each user. Related works indicate phrases and UIs of dialog messages and incentives that influence a user’s action. In our previous work, we focused on smartphone users updating software, and proposed appropriate phrases of dialog messages according to the user’s understanding of the updating procedure, as well as the type of software. We also analyzed appropriate incentives. However, in the terms of level of literacy, the effectiveness of the UI of dialog messages and the volume of incentives remain unclear. In this paper, we conducted a user survey to analyze appropriate UIs according to the user’s understanding of the updating procedure and the appropriate volume of incentives. As a result, we confirmed different UIs are effective according to the user’s understanding of the updating procedure. In addition, we found an appropriate volume of points, mobile data, and coupons in order to promote the updating of software.

This paper provides a solution to mitigate mistrust that arises from untransparency in a multi-stakeholder environment. This work is related to a kind of verifiable computation. When considering the multi-stakeholder environment, not only the participants’ requirements but also the assignee’s intention should be respected. That is, the assignee should be given the discretion to select a result that is the best for the assignee among all the choices. However, there is a possibility that if the assignee is malicious, he/she may falsify and provide an inadequate result for participants to maximize his/her benefit by ignoring the participants’ requirements. It is difficult for the participants to detect this if they want to keep their preference orders secret from others. This paper proposes a solution to determine whether the received result is adequate for the participants while keeping their preference orders secret. The proposed solution is based on fully homomorphic encryption (FHE) and assumes the use of a semi-honest third-party server. This paper first describes a general solution that is not limited to specific requirements from participants. Next, this paper shows a way to apply to stable matching problem as a specific implementation. More specifically, a transformation of a standard stability checking algorithm into an algorithm that can be implemented by FHE with the computational complexity O(n2)$O(n^2)$. Finally, this paper gives an example of an implementation and its performance with HElib, which is an FHE library that provides BGV.

The local differential privacy metric has attracted attention due to its quantitative nature, and many mechanisms have been studied for satisfying local differential privacy based on data formats and use cases. Local differential privacy mechanisms generally target a certain data space and perturb it sufficiently to provide indistinguishability of the data on that space. Therefore, individual data tends to be greatly disturbed so that even relatively simple tasks require a large amount of data to equalize the noise caused by the mechanism. In this paper, we define hierarchical local differential privacy, which is an extension of local differential privacy, and propose a mechanism to satisfy both local differential privacy and hierarchical local differential privacy. Hierarchical local differential privacy views a data space hierarchically as a set of smaller spaces, and instead of abandoning the privacy of data contained in different spaces, the amount of noise can be reduced. In this paper, we further design a hierarchical local differential privacy framework and achieve a privacy guarantee based on local differential privacy for all the data in the framework. Finally, we experimentally evaluate the proposed framework using image data. The framework allows control over the amount of information that can be disclosed, and furthermore, maintains a higher degree of utility than applying a simple local differential privacy mechanism.