Results of the FY2023 MITOU Target Program: Software Development Utilizing Reservoir Computing Technology

In-page Links

Results of the FY2023 MITOU Target Program: Software Development Utilizing Reservoir Computing Technology

PM: KATORI Yuichi

Development of real-time visualization and analysis software for sensor data in offline environment using reservoir computing

The “CanteenFlow” software was developed to utilize reservoir computing to perform real-time visualization and analysis of sensor data in an offline environment. The software runs on a web browser and can be connected to sensor devices. Once data is received, it is intuitively displayed as a graph. Received data can also be automatically classified and used for learning. The software is implemented as a progressive web app and is easy to install and update. The reservoir computing feature allows visualization, analysis, and learning of sensor data while conserving computational resources. For researchers and corporate engineers, efficient research activities and social implementation are expected.

• View the details of Development of real-time visualization and analysis software for sensor data in offline environment using reservoir computing (in Japanese)

Creator: KUROTAKI Yuta

Construction and implementation of a reservoir system using quantum computers for prediction of stochastic processes

Quantum reservoir system is a machine learning model that uses quantum computers for time-series data prediction. In this project, we developed a method for predicting time-series data with noise using the quantum reservoir system and made it widely accessible to the public as a web application. Since quantum computers provide outputs as probability distributions, they are expected to effectively predict time-series data with noise. In this project, we designed and benchmarked the quantum reservoir system for noisy time-series data and implemented it as a web application. The novelty and superiority of this project lie in handling stochastic processes (noisy time-series) with reservoir computing and utilizing this system on real quantum computers. In the future IoT society, where predicting large-scale data with noise will become increasingly important, this project is expected to make a significant impact on society.

• View the details of Construction and implementation of a reservoir system using quantum computers for prediction of stochastic processes (in Japanese)

Creators: YASUDA Toshiki, HISHIDA Kaisei, KAMEI Masato

PM: TANAKA Gohei

RiP: High-Efficiency Reservoir-in-Processor Computing Utilizing Internal State of RISC-V Processor

Reservoir computing is a machine learning technique that enables fast learning with low power consumption. However, existing reservoir computing implementations require processors for matrix operations and reservoir control, as well as a physical system to realize the reservoir, making it difficult and costly to construct a system. Moreover, the communication between the reservoir and the processor impacts the latency of the computation. This project aims to develop a highly efficient reservoir computing architecture that utilizes the internal state of a processor. We evaluate the performance of the proposed processor system by running benchmarks on it realized on FPGA. The results of this project are expected to open up a new research field and provide a foundation for it. Furthermore, we make the artifacts open source, and they are expected to contribute to industrial applications as well as the RISC-V community.

• View the details of RiP: High-Efficiency Reservoir-in-Processor Computing Utilizing Internal State of RISC-V Processor (in Japanese)

Creators: ASANO Kohei, SUGA Kengo, TANAKA Sun, HIRAYAMA Yuki

Development of a Personalized Emotion Estimation Application Using Heart Rate Data

This project developed EmoNote, a personalized emotion estimation application using heart rate data. Emotions are an important factor affecting daily life, and conventional emotion analysis methods have been unable to capture individual emotions and deeper feelings. In this project, we will develop an emotion analysis technology based on heart rate data to estimate emotions for various situations in users' daily lives. By doing so, the project aims to capture the subtle nuances of emotions and improve the practicality and universality of people's emotion recognition. We also adopt a local approach to data processing, taking into account privacy issues and security risks. The results of this project are expected to be applied not only to improve the mental wellbeing of individuals, but also in a variety of other areas such as healthcare, education, and the workplace. It is also expected to contribute to the promotion of a culture of emotional understanding and empathy.

• View the details of Development of a Personalized Emotion Estimation Application Using Heart Rate Data (in Japanese)

Creators: FUKUHARA Rikuto, KIDA Takuma