Soramichi Akiyama

Hardening Operating Systems against Side-channel Attacks by Concealing Data Addresses

Researcher

Soramichi Akiyama

Outline

Memory side-channel attacks have been proven to enable reading/overwriting sensitive information that are normally not accessible only by executing a program on the target computer. These attacks require (1) that an attacker and the victim share data location in the storage and (2) that the shared data location can be found from data addresses. The goal of this research project is to mitigate risks of memory side-channel attacks fundamentally by blocking the requirement (2) based on ideas of separating memory management from the OS and isolating it in an trusted execution environment.

Kota Ando

Cooperative AI training platform for user-oriented data utilization

Researcher

Kota Ando

Outline

This project aims at establishing a user-oriented AI platform where the users can construct and train neural network (NN) models using their own data. Training an AI model at this time is highly cloud-based due to its expensive computational requirements and need for massive training datasets. Motivated by federated training techniques, we propose a training algorithm that shares a part of the models under training with users with different interests and private data. We present a lightweight processor with an aggressive approximation and stochastic computing nature. These in combination enable user-side NN training utilizing local personal data and computational resources.

Yutaro Kashiwa

Developing Technique for Anomaly Detection in Software Behaviors

Researcher

Yutaro Kashiwa

Outline

This project aims at establishing a technique for detecting anomalies and defects without using developer-created tests. The technique specifically makes use of dynamic software analysis methods to create trace logs － that is, records of the lines of code that are executed － while manipulating the software products before and after changes. The technique then quantifies the generated trace logs (i.e., their behaviors), checks significant differences between the behaviors before and after changes, and detects which of them are anomalies. The technique also checks developers’ commit messages to see if the behavior changes are expected by developers (i.e., context-aware anomaly detection).

Bo Wei

Intelligent network management with integrated communication and sensing

Researcher

Bo Wei

Outline

mmWave is the next-generation technology which can enable the high-speed communication. However, it is easy to be blocked by obstacles. On the other hand, the wavelength of mmWave is short which is promising for high-accuracy sensing. In this project, the integrated communication and sensing (ICS) technology as well as the software platform will be developed. By taking advantage of the physical information obtained from ICS, the intelligent network management technology will be developed which is secure, privacy-protected, and highly reliable. This project will contribute to the autonomous driving, telehealth, smart city, and the realization of Society 5.0.

Takuya Kojima

An innovative reconfigurable system design for confidentiality and integrity protection

Researcher

Takuya Kojima

Outline

This research is based on a SW/HW co-design approach to guarantee the confidentiality and integrity of information for next-generation computer systems with reconfigurable hardware, which is promising in terms of energy efficiency and processing performance. Primarily this research project aims to produce tamper-proof hardware configuration information, tamper-resistant techniques against side-channel attacks, and design methods for high-performance computing systems to make fully homomorphic encryption practically feasible for secure computing.

Hideaki Hata

Social implementation of meta-maintenance to maintain software ecosystems

Researcher

Hideaki Hata

Outline

Toward the social implementation of meta-maintenance for comprehensive maintenance of open source software ecosystems, this project will conduct large-scale empirical analyses to search for useful knowledge from open source software development projects, develop systems for identifying useful knowledge and monitoring open source software ecosystems, and develop human-in-the-loop technologies and mechanisms for distributing the identified and aggregated knowledge to the entire open source software ecosystems.

Daichi Fujiki

Enabling Safe and Efficient Processing of Data by Memory-Centric Database System

Researcher

Daichi Fujiki

Outline

In this research, we develop a PIM/IMC-empowered database system that allows the read efficiency of the B-trees and the update efficiency of the LSM-trees to coexist for HTAP workloads. We also enable sustainable and efficient processing of privacy-preserved databases that use homomorphic encryption to protect data and user queries. By introducing IMP, we will transform passive memory resources near the static data for active compute elements, which will significantly contribute to minimizing data movement.

Taiki Miyanishi

Developing City-Scale Navigation AI that Considers Potential Risks

Researcher

Taiki Miyanishi

Outline

This project aims to realize a navigation AI that can follow the user’s instructions and suggest safe routes to a mobile agent even under challenging situations by: 1. Creating city-scale 3D maps and automatically adding semantic information. 2. Developing a language grounding method for the 3D map. 3. Developing a navigation AI that follows linguistic instructions. 4. Improving the navigation AI to be robust by using disaster and accident risk simulation.

Shinobu Miwa

A Framework to Support Use of Trusted Execution Environment in High Performance Computing

Researcher

Shinobu Miwa

Outline

In this study, I develop an LLVM-based code transformation framework for High Performance Computing (HPC) users that requires fast and secure execution of their applications. The proposed framework enables to automatically extract processes needed for Trusted Execution Environments (TEEs) from parallel applications written for Rich Execution Environments (REEs) and then generate codes running on REEs and TEEs concurrently while performing high-speed encrypted communication between multiple TEEs. The proposed framework will be validated with various parallel applications including real HPC applications.

Kentaro Yoshioka

A Sensor Security Research Across Cyber and Physical Domains

Researcher

Kentaro Yoshioka

Outline

Sensor spoofing attacks that exploit vulnerabilities in LiDAR sensors can be a serious threat to self-driving vehicles and can trigger accidents by such attacks. In order to protect self-driving cars from the threat of sensor spoofing, we conduct research on sensor security across cyber and physical domains. We will create new security technologies based on sensor and software co-design, and foster a research environment that enables early detection of vulnerabilities by encouraging information sharing between cyber and physical domains.