*This Program finished in FY2021
　Active Period: FY2016 - FY2021
　Affiliations and titles are at the end of each research activity.

Soramichi Akiyama

Auto-Acceleration of Applications on Next-generation Memory Devices

Researcher

Soramichi Akiyama

Outline

Data IO to/from memory is much slower than computation in CPUs, resulting in poor performance of many applications. This research tackles this issue by leveraging a type of next-generation memory devices that allows faster accesses in return of some errors (bit-flips) injected into the data. We focus on hardware/software mechanisms that estimates where errors occur inside memory and how these errors affect the output quality of given applications in order to guarantee the output quality of the applications under given requirements.

Shinji Ito

Online Convex Optimization Based on Partial Feedback

Researcher

Shinji Ito

Outline

Though technologies of online convex optimization are useful for sequential decision making, they cannot be applied directly to real-world problems in which only partial feedback regarding objective functions are available. In this research project, we consider online convex optimization with partial feedback, aiming to develop optimal algorithms and to prove their optimality.

Danilo Vargas

Automatic Search for Robust Hybrid Deep Learning Models

Researcher

Danilo Vargas

Outline

In my ACT-I’s research output, some features of intelligent systems which made them more robust than others were identified. Among the highlighted features present in robust intelligent systems there are (1) the capacity of learning high level concept and (2) dynamic routing and feedback mechanisms. In the Kasoku Phase of this project, novel intelligent systems possessing (1) and (2) will be investigated, walking towards to a solution to adversarial machine learning.

Taihei Oki

Exploring theory and applications at the intersection of combinatorial optimization and linear algebra

Researcher

Taihei Oki

Outline

Combinatorial optimization and linear algebra are closely interrelated, and theoretical tools in one field are often used to solve problems arising in another area. In this project, we aim to deepen the understanding of “algebraic combination optimization,” which is the intersection of these two fields, from both theoretical and practical aspects. In particular, we will develop (i) a nonlinear extension of the correspondence between polynomial matrices and combinatorial optimization problems, and (ii) applications of combinatorial optimization theory to systems analysis.

Yuta Suzuki

Modality transformation for materials measurement data

Researcher

Yuta Suzuki

Outline

In modern material development, analysis of measurement data is one of the bottlenecks in the research workflow. In this research, we develop a rapid data analysis method for various modal data; it based on the idea that the estimation of materials properties from measured data can be regarded as a modality transformation task in machine learning. Furthermore, with combining machine learning prediction and physical models, we aim to ensure not only rapid prediction but also physical validity.

Yugo Nakamura

Next-generation nudge based on context recognition using various IoT devices

Researcher

Yugo Nakamura

Outline

Nudge that leads people into desired behavior without forbidding any options is expected to be applied in a wide range of fields. However, the effectiveness of existing nudge methods does not persistent because of the limitation when adopting a static and one-pattern information presentation/ intervention under a specific situation. In this research, we aim to develop the next-generation nudge that has more efficient and effective. It continuously intervenes at appropriate timing and manner based on the behavioral context recognition method using various IoT devices in the living environment.

Yu Nakayama

Adaptive Reconstruction of Telecommunication Network

Researcher

Yu Nakayama

Outline

Spatio-temporal patterns of human mobility result in significant fluctuations of mobile traffic. Such fluctuations drastically deteriorate the efficiency and financial viability of conventional mobile networks. To address this issue, this research investigates an adaptive mobile network architecture with moving nodes towards beyond 5G era. This work also targets network architectures and utilization methods based on crowdsourcing and sharing economy to further improve the adaptivity.

Koya Narumi

Digital fabrication of biomimetic user interfaces

Researcher

Koya Narumi

Outline

This project aims to achieve biomimetic interfaces that can, for example, actuate, sense, communicate, harvest, or heal by themselves. Especially, it will explore the self-actuating mechanism derived from programmable metamorphosis of living things, and apply this mechanism to real-world interfaces.

Hideaki Hayashi

Probabilistic generative model of biosignals and neural networks for inference

Researcher

Hideaki Hayashi

Outline

In this study, we develop probabilistic generative models that describe the generation process of biosignals. We also propose neural networks based on probabilistic models for estimating such complex probabilistic generative models. The application of this study involves medical diagnosis support and human-machine interfaces using biosignals. Furthermore, this study leads to the development of a framework for integrating probabilistic generative models and neural networks.

Satoshi Funabashi

Realization of Tasks by Multi-Fingered Hand Using Machine Learning and Distributed Tactile Sensors

Researcher

Satoshi Funabashi

Outline

In order for robot hands to handle tools and objects, it is necessary to achieve various in-hand manipulation tasks with complicated contacts with fingers and a palm. Conventionally, CNNs have been used to process the tactile information. However, it cannot adapt to the complexity of the positional relationship between fingers and a palm, changes in spatial sensor arrangements due to the changes in joint angles during in-hand manipulations. In addition, it is necessary to recognize the object characteristics such as the softness and weight of various objects to adjust the manipulation motion by reflecting it on the manipulation of multiple fingers. In this study, we aimed at achieving multi-fingered in-hand manipulation tasks by adapting to these task specific problems.

Yuto Miyatake

Uncertainty quantification in estimating ODE models

Researcher

Yuto Miyatake

Outline

When estimating parameters of differential equations, uncertainty indued by using numerical solutions could have substantial impacts. The study of ACT-I has developed a prototype method of quantifying the uncertainty. This study aims to verify the method mathematically, to generalize the method to a variety of inference methods such as Bayesian, to extend the method such that it can deal with whole uncertainty in the estimation process, and to apply the method to large scale practical problems.

Tatsuya Yokota

Multiway Delay Embedding and Modeling of Tensors

Researcher

Tatsuya Yokota

Outline

In this study, we propose a new framework of tensor modeling using multi-way delay-embedding which is an extension of delay-embedding of time-series. We develop reconstruction, inference, and control methods which utilizes locality and self-similarity of tensors. Our goal is to provide various signal processing applications such as for audio, image, video, medical, remote sensing, and robots.