Keisuke Okumura

Environment Optimization for Multi-Agent Path Planning

Researcher

Keisuke Okumura

Outline

This project studies the optimization of environments for multi-agent path planning, wherein the environments are designed by obstacle layouts or installing one-way roads, to name just a few. By automating environmental design that conventionally has been manually adjusted, the project aims at providing efficient and safe environments for multi-robot coordination. Doing so is expected to be beneficial for several important domains including factory automation. Developed methodologies will be tested by both simulation and real-robot demonstrations as proof of concept.

Ryusei Oketani

Chirality recognition of molecular crystals by deep learning

Researcher

Ryusei Oketani

Outline

This project aims to develop a deep learning model enabling chirality recognition of chiral single crystals from microscopic images for facile and rapid measurement of enantiomeric excess in solid state. The chirality of crystalline solids has been characterized by either spectroscopic method or X-ray diffraction measurement so far. However, these techniques need a time-consuming data collection to obtain information on anisotropy or atomic coordinates in crystals. Toward a facile and rapid measurement of chirality of crystals, a deep learning model in end-to-end fashion, i.e., from crystal detection to chirality recognition, will be constructed. This work would contribute as a novel enantiomeric excess measurement of crystalline materials.

Jun Kato

AI-enabled Creativity Support Environments for a Constructive Study of Creative Culture

Researcher

Jun Kato

Outline

To realize a creative society in which people want to create content with the help of AI technologies, a sustainable ecosystem in which content inspires the next creation serves a critical role. We shed light on the creative culture in such an ecosystem, build a “creativity support environment” that supports the content creation and distribution process, and aim to establish Human-Computer Interaction technologies that engineer, nurture, and maintain the creative culture.

Seiya Kawano

Patent Discourse Structure Parsing for Assisting Patent Document Reading

Researcher

Seiya Kawano

Outline

This project aims to realize a discourse structure parsing model that identifies the structure of patent documents at a fine-grained level, and provides an analytical foundation for understanding the technological scope of inventions.

Shun Kojima

Deep Learning for Integrated Wireless Communications and Radar Signal Processing

Researcher

Shun Kojima

Outline

With the remarkable development of IoT technology, in addition to the increasing demand for wireless communications, radar applications of radio waves are gaining attention as important social infrastructure. Against this background, the frequency bands available for wireless communications and radar are tightened, and spectrum congestion is an urgent issue. In this study, we focus on the waveform format of signals and introduce a deep learning structure suited to the characteristics of wireless signals to create a new method that enables both high-speed wireless communications and high-precision radar signal processing.

Risa Kobayashi

Mutual Dynamization of Land Holding Theory Based on Very Long-Term Observations and Its Renewal

Researcher

Risa Kobayashi

Outline

In this study, I attempt to automatically digitize all items of the real estate registry and construct a behavioral model that takes into account observation errors. I work on detecting the layout of the digitized real estate registration documents and recognizing the characters in the detected text boxes. Next, I attempt modeling using a general state space model. The model applies a learning model that learns the error between the vector of data observed by layout detection and character recognition and the state vector estimated by the system model.

Mizuka Komatsu

Modeling Auditory Attention: Coordinating a Dynamic Causal Model and Deep Learning

Researcher

Mizuka Komatsu

Outline

The ability to select and focus on a part of sound from the outside world is called auditory attention. Existing models of auditory attention can be divided into two broad categories: the bottom-up approach, which focuses on the sound itself; and the top-down approach, which focuses on listening tasks. However, advances in data measurement technology hold the promise of more advanced modeling methods capable of integrating both types of models. This study pursues this possibility by coordinating mathematical and information sciences and their technologies, such as dynamic causal modeling and deep learning.

Ryuta Shiraki

Intelligent photonic networks exploiting multidimensional big data

Researcher

Ryuta Shiraki

Outline

To realize photonic networks that flexibly offer large-capacity communication services, we develop an intelligent networking method that exploits multidimensional big data. We tackle this problem with two key technologies; the optical monitoring technology and optical controlling technology. First, the informative features for controlling networks are extracted from the various data obtained by advanced optical orchestrating technologies. Then, the photonic network is intellectually controlled by using the sophisticated machine learning algorithms under the constraints specific to the photonic network.

Yuichiro Sueoka

Innovative Swarm Robot System Science Based on Self-Attention Neural Network

Researcher

Yuichiro Sueoka

Outline

In this research, we present an innovative swarm robot system science that provides unprecedented robustness, flexibility, and scalability to swarm robot systems by applying the natural language processing technology: Transformer. Specifically, for the problem of cooperative transportation of objects by swarm robots, we will design a neural network that can respond to changes in the number of robots and in the number and mass of objects to be transported, and test the validity of the proposed system through verification on actual equipment.

Sato Naito

Creating of systems for the use of “trusted AI” in the public sector

Researcher

Sato Naito

Outline

Many Japanese local governments are considering starting to use AI. However, there are many indications by legal and AI experts that the use of AI in the public sector can infringe on human rights. Denmark, one of the most digitally advanced nations in the world, is using AI in the public sector while respecting human rights and transparency. This study will identify which technical and legal systems of AI used in the public sector in Denmark, the comparative law subject of this study, can serve as a reference for Japan.

Toshiyuki Nakanishi

Development of a prediction AI model of postoperative pain exacerbation for automated analgesia

Researcher

Toshiyuki Nakanishi

Outline

This project aims to develop an AI for predicting postoperative pain exacerbation in real-time. Original time-series datasets will be constructed using a newly developed continuous pain assessment method and biosignal measurements on patients undergoing surgery. An anomaly detection algorithm based on a deep learning model capable of processing time-series data will be adopted in the developing AI model. Furthermore, we will implement the developed AI model as a smartphone app to verify its usefulness in clinical settings.

Masashi Hamaya

Learning robotic assembly using a soft wrist and tactile sensors

Researcher

Masashi Hamaya

Outline

This project addresses learning control for robotic industrial assembly under uncertain environments and unseen parts. The key point is to leverage a soft wrist that allows safe physical interactions with the environment and tactile sensors that can observe the contact events for subtask achievements during assembly. The real robot experiments will reveal that the proposed method can complete the insertion tasks even with unseen parts and grasping positions.

Daichi Haraguchi

Creation of visual design informatics

Researcher

Daichi Haraguchi

Outline

Designers create various visual designs (fonts, logos, posters, etc.) based on implicit and empirical knowledge. An example that we can observe in our daily life is that red tends to be used to express the impression of “warm.” This project aims to quantify designers’ implicit and empirical knowledge by applying image analysis technology and machine learning technology to visual design.

Haruna Fukushige

Creating a predictive model for nurse call occurrence that considers personal characteristics.

Researcher

Haruna Fukushige

Outline

To provide high quality nursing care, it is important that nurses understand the personal characteristics of patients, and are allotted to sufficient time for care. In this study, we will create a predictive model for nurse call occurrence. Because nurse call is an important factor in the performance of nursing tasks.This model will provide a more effective and efficient approach to responding to nurse calls. The purpose of this study is to provide more personalized nursing care to patients and to gain time for nurses to provide care.

Yin Kan Phua

Creation of AI-driven Functional Polymer Informatics

Researcher

Yin Kan Phua

Outline

Functional polymers are fundamental materials that are used as key components in fuel cells, batteries and so on. This study first aims on building a data-centric explainable machine learning model by utilizing graph neural network and attention mechanism to functional polymers used in fuel cells. Besides predicting the target variable, prediction logic of this model will also be interpretable to chemists. The ultimate goal of this study is to establish functional polymer informatics, which consists of an explainable machine learning model building methodology that is generalizable to any other functional polymers besides those of fuel cells. Generalization of the method developed in this study is believed to help accelerate and cut down cost of researching and developing functional polymers.

Keigo Matsumoto

Estimation of sensory characteristics and optimization of sensory information presentation based on the free-energy principle

Researcher

Keigo Matsumoto

Outline

In this research, the sensory characteristics of each individual are estimated using a sensory characteristics estimation model based on the free-energy principle, using the hidden state of the virtual environment and observation data obtained from psychological, behavioral, physiological, and brain function measurements, to optimize the presentation of sensory information to each individual. This research is expected to lead to the realization of interfaces that optimize and present sensory information individually in virtual environments, as well as to the solution of problems related to sensory characteristics such as hypersensitivity and hyposensitivity in the real world.

Takahiro Miura

Quantitative Understanding of Knowledge Gap in Research Toward Paper Recommendation

Researcher

Takahiro Miura

Outline

This project aims to build a model to identify knowledge gap areas in research by focusing on delayed recognition in science. In general, cutting-edge research appears from the surroundings of popular fields, known as the ‘bandwagon effect.’ Empirically, the knowledge gap functions complementarily, causing disconnection and discord in science to enhance the non-linear growth of research fields, but the general application for researchers utilizing the knowledge gap is still in its infancy. To overcome this, this project will develop the model to recommend papers falling into the knowledge gap for the input papers, with the goal of implementing the application to stimulate researchers’ emerging knowledge creation.

Takahiro Mimori

Development of an unknown-aware AI system for blood cell analysis

Researcher

Takahiro Mimori

Outline

The construction of AI system to automatically differentiate blood cells in peripheral blood and bone marrows is an important task, but it involves difficulties in accurately annotating a wide variety of morphologies and collecting data in diverse imaging environments. This project aims to develop an AI method that can recognize unknown condition of cells and build a technology that enables screening without assuming detailed classification by human hands. We will also develop interpretability techniques for effective analysis and methods that are robust to environmental changes.

Takuma Yagi

Visual Script Prediction based on Knowledge Extraction from Large Language Models

Researcher

Takuma Yagi

Outline

I will develop a visual script prediction model which predicts scripts (typical event sequence) from images. While current image recognition models are good at recognizing visible objects in the scene, they are still far behind humans in recognizing their invisible contexts. In this project, I aim to realize an AI that can take appropriate actions without detailed human instructions by extracting script knowledge from large-scale language models that are said to include diverse contextual knowledge.

Ryo Yamaguchi

Establishment of eco-evolutionary experiment controlled by Real-Time AI

Researcher

Ryo Yamaguchi

Outline

In the life sciences, populations/cell populations are highly heterogeneous, and their interactions are complex. Thus, predicting future dynamics involves various difficulties. I consider a continuous yeast culture system as an artificial ecosystem in this research. And aim to develop a platform to automatically control complex life phenomena, such as the evolution of resistance mutations to drugs and other toxic environments, and the coexistence of several competing species, through a combination of real-time time series data aquisition and AI.

Takahiro Yamazaki

Iron-loss analysis of nanocrystalline magnetostrictive composite using magnetic Barkhausen noise

Researcher

Takahiro Yamazaki

Outline

This project aims to elucidate the iron-loss mechanism that determines the energy conversion efficiency in electric motors by proposing a new magnetic Barkhausen noise analysis method based on magnetic physics and information science. Specifically, the relationship between nanostructures and iron loss phenomena through magnetostriction will be discussed. Furthermore, new material design to achieve higher motor efficiency will be established by formulating the iron loss factors visualized by machine learning on a physics-based feature space.

Gaku Yokoyama

Development of a hybrid photosynthetic model for accelerating advances in agricultural production technology

Researcher

Gaku Yokoyama

Outline

Photosynthesis is an essential physiological process for agricultural production. As agricultural production becoming unstable due to frequent extreme weather events under climate change, it is necessary to develop a model that can predict photosynthetic responses to future environment. In this project, we aim to develop a hybrid photosynthetic model for field crops by combining the biochemical photosynthetic model and ANN model, which is easy to construct but has high explanatory and extrapolative power.

Hikaru Yokoyama

AI-based non-invasive investigation of neural activity of deep tissues during whole body movements: Pushing the limits of electromyography and electroencephalography

Researcher

Hikaru Yokoyama

Outline

Applying convolutional neural networks to electroencephalography (EEG) and electromyography (EMG), I will achieve a non-invasive estimation of activities of the deep brain and spinal cord neurons during whole-body movements. Furthermore, to demonstrate the usefulness of the proposed method for understanding motor control, I will elucidate the cortical-deep brain-spinal cord network for gait control by the proposed method. This will open up a measurement area that has been unreachable for about 100 years since the development of EEG and EMG devices and will bring innovation to a wide range of research fields in which these devices are used.