AIP Network Lab

  • Font-size
  • Large
  • Middle
  • Small

List of AIP Challenge PRISM Acceleration Support

Constructing a multiscale infectious disease transmission model and its application to outbreak prediction

Yusuke Asai (Assistant Professor,Graduate School of Medicine,Hokkaido University)

Infectious diseases such as influenza, SARS and HIV spread through human-to-human contact. Their force of infection depends on the amount of viruses in infected individuals, however, the virus dynamics is not uniform because of the heterogeneity of infected population. In this study, virus dynamics and population dynamics are combined and a multiscale infectious disease transmission model is constructed. Evaluation of the impact of virus scale changes on human level transmission and identifycation of the contributing factors allow us to develop health countermeasures and prevent further disease transmission.

Correlations between father-child interacton and brain structure

Michiko Asano (Postdoctoral Researcher,Graduate School of Arts and Sciences,The University of Tokyo)

Many of the existing researches on children focused on the interaction between mother and the child, and there was little that took up that took up the interaction between father and the child. Some studies on ASD children have reported that response and encouragement not only from the mother but also father is important for children to obtain sociocentricy, suggesting the importance of the father and child relationship. This study examines fathers of typically developed child and child with ASD, evaluates the quality of interaction between the father and the child, and then clarify the relationship between the quality and the neural base.

High throughput method for validation of NMR protein structures and accurate predictions of allostery

Adnan SLJOKA (Assistant Professor,Department of Informatics School of Science and Technology,Kwansei Gakuin University)

Accurate picture of proteins atomic structure is important to understand their biological function. Structures are usually solved by X-ray crystallography or with Nuclear Magnetic Resonance (NMR) experiments and deposited into Protein Data Bank (PDB). Unlike X-ray structures, the quality of NMR protein structures is difficult to validate. Previously, we showed we can use Random Coil Index (RCI) method together with our FIRST-ensemble method to assess the quality of NMR structures. We will further apply this to create a data set of high quality NMR structures in PDB, identify poor members of NMR structural ensemble with a final goal to create a server for NMR validation process and incorporation to PDB. This will for first time will allow us to assess the accuracy of NMR structures and improve structure calculation process. Another aspect of the project involves using identified high quality structures for predictions of allostery using our rigidity transmission allostery RTA (as appeared in Science 2017) method.

A Probabilistic Framework for Unsupervised Grounding of Combinatory Grammatical Structure of phrases in Situated Human-Robot Interaction

Amir Aly (Senior Researcher,The Research Organization of Science and Technology,Ritsumeikan University)

In this project, I would build an integral framework to infer the latent grammatical structure of language, which embraces grounding Parts of Speech, and induction and grounding of Combinatory Categorial Grammar (CCG) - that also allows for interfacing syntactic and semantic representations of language - through visual perception. This could help in making a robot able to understand syntactic relationships between words (i.e., understand phrases composing a sentence) in addition to semantic representations of words and phrases, which consequently leads to understanding the meaning of human instructions during interaction.

Estimation of effect to supply chain networks during an earthquake tsunami disaster

YOSHIKI OGAWA (Project resercher,Institute of Industrial Science,The University of Tokyo)

This research develop a method to simulate how economic damages to intercompany transactions will spread and recovary through supply chain (SC) during the Nankai Trough Earthquake by using various big data such as bussiness transaction data and mobile phone GPS data. For that purpose, estimating the damage of each company by the Nankai Trough Earthquake Tsunami and set model of damage spread after the disaster to the recovery stage and simulate throughout the SC. Finally, we clarify the effect of the industrial location structure and how effect the differarece of policy decision making on the SC recovery.

Automatic Diagnosis by Utterance Analysis of Children with ASD

Masahito Sakishita (Student (Master's course),Graduate School of Integrated Science and Technology,Shizuoka University)

Autism spectrum disorder (ASD) is a common disorder with a prevalence of 1 to 2%. This diagnosis can waver depending on circumstances of patients at the time of the medical examination, clinicians, and the environment. I extend an ASD utterance corpus that I have constructed, in order to predict severity of children with ASD and to classify them into typical development or ASD. In addition, I improve efficiency in a transcription by speech recognition. Through this study, I aim to find speech features that potentially characterize ASD. This study will be a foothold for support of clinical screening process and language rehabilitation.

Heart Rate Estimation Based on Markerless 3D Pose Position

Hidehiko Shishido (Assistant professor,Center for Computational Sciences,University of Tsukuba)

In this research, we devise a heart rate estimation algorithm based on markerless 3D pose position for sports training support. Heart rate estimation from sports video is realized by machine learning the relevance between the 3D pose position and the heart rate using the Result (Japanese Only)s so far. And we aim to accelerate the practical application of research Result (Japanese Only)s. According to this method, the heart rate is estimated without attaching the heart rate measuring instrument. Therefore, it is expected to be applied to situations such as games where attaching of heart rate measuring instruments is impossible.

Developing Data Science Methodology for Precision Medicine

Shonosuke Sugasawa (Lecturer,Center for Spatial Information Science,The University of Tokyo)

There is a growing demand of personalized medicine that aims to provide adaptive medical treatment according to the characteristics of patients, and analysis of medical big data plays a central roll to achieve the goal. In this project, we develop data science methodology that can find valuable knowledge based on large amounts of genetic information. In particular, we employ recent machine learning techniques to develop effective methods to identify genetic information that are strongly associated with heterogeneous treatment effects.

Automatic measurement and developmental support system of social interaction

Satoru Sekine (Graduate Student(Doctor's course),Graduate School of Human Relations,Keio University Graduate School)

I will implement the early development support program "Movement-Interaction Teaching" developed by me. To that end, we develop a sensing device and a protocol that advances support based on real-time data. In particular, I impement automatic measurement and feedback of social interaction between two parties (eye contact, approach, follow-up, synchronization). I set up a development support platform named "Movement-Interaction Teaching" and collaborate with engineering field led by psychology.

Develpopent of analysis for dynamic changes in task-related brain networks which were correlated to our personarity and socio-economic status

Toshiko Tanaka (Resercher,Center for Information and Neural Network,National Instutute of Information and Communitations Technology

The link between the functional connectivities under the resting state and individual personality traits has been indicated. However the relationships between the task-related dynamic change in the functional network and our trait remain unclear. We try to develop the optimal way for analyzing the task-related dynamic changes in functional networks and apply it to the investigation of the relationships between the pattern of dynamic network change and personality.

Path planning from speech commands based on stochastic reasoning using spatial concepts

Akira Taniguchi (Project resercher[Research Fellow of Japan Society for the Promotion of Science (PD)],Research Organization of Science and Technology,Ritsumeikan University)

In this research, we aim to enable navigation using spatial concepts in a mobile robot through the mathematical integration of reinforcement learning to a stochastic reasoning problem on the probabilistic generative model. We develop a method to use concepts formed based on the robot's own experience by unsupervised learning. Specifically, we realize the path planning to the target state of the spatial concept estimated from the speech command of the user such as "go to the kitchen". We also perform navigation tasks through human-robot interaction and verify effectiveness.

Evaluation of Face-to-face in Workshop Using VR Immersion

Tomohiro Nishida (Specially-appointed Assistant professor,Center for Collective Inteligence,Graduate School of Engineering,Nagoya Institute of Technology)

When participants consider ideas alone in workshop, we found that VR immersion of walking by head-mounted display is effective. However, there was a problem that the discussion became difficult. It is thought that the face-to-face is obstructed.Therefore, in this study, we aim to clarify face-to-face and psychological / physiological effects in workshop using VR immersion. This study aim to utilize for workshop design using VR immersion.

Study on Cultural Diversity of Interpersonal Touch Effect on Social Behavior

Taku Hachisu (Researcher,Faculty of Engineering,Information and Systems,University of Tsukuba)

The purpose of this study is to construct an experimental field in the United State by conducting a research on interpersonal touch quantification and facilitation. This would allow for comparing the interpersonal touch behavior with that in Japan and for understanding the cultural diversity of interpersonal touch effects on social interactions. To this end, the researcher first asks collaborators in the United States to use wearable devices that the researcher has developed to quantify and facilitate interpersonal touch. Then, the researcher investigates usability and user experience and improves the performance of the device to make the future field experiment efficient.

Missing value imputations by intervals

Hiroyuki Hanada (Project Assistant Professor,Graduate School of Engineering,Department of Computer Science,Nagoya Institute of Technology)

Many machine learning methods assumes that the dataset has no missing value, but it is not always the case for real-life datasets. As a better approach than assigning one value to each missing value (including methods trying multiple cases of assignments), we consider assigning an interval; which Result (Japanese Only)s in the predicted values being also given as intervals. For example, if we do not know a person's height, we replace it with an interval that includes values likely to occur (e.g. 130cm to 180cm).

Human-Computer Interaction Framework for Assisting Context-dependent Decision Making

Keita Higuchi (Project Research Associate,Institute of Industrial Science,The University of Tokyo)

This study proposes a human-computer interaction (HCI) framework to assist context-dependent decision-making through the use of an online learning approach for discovering user preferences. The proposed approach bootstraps the user preference model with a set of policies that are learned with supervised learning from a dataset collected from other users. Then the online learning approach quickly adapts to a target user by learning and tracking the optimal policy that best describes the target user's actions. We use the proposed framework to address two concrete interaction scenarios of automated navigation and image filter selection.

Developing the approximation method based on variational upper bound using non-equilibrium physics

Futoshi Futami (Graduate Student(Doctor's course),Department of Complexity Science and Engineering,Graduate School of Frontier Sciences,The University of Tokyo)

For the practical use of machine learning, it is very important to estimate how much certainty the algorithm can certify on its output.In Bayesian inference, such certainty can be considered based on the model likelihood for test data. An existing approximation method overestimates the likelihood and thus Result (Japanese Only)s in trusting the algorithm excessively, and it suffers a huge calculation cost. In thisresearch proposal, we develop a method that does not overestimate the likelihood while suppressing the computational cost by using the method of nonequilibrium physics.

Development of Wearable Interface that Leverages Touch Gestures to Face

Katsutoshi Masai (Project Research Associate,Graduate School of Science and Technology,Keio University)

In this research, we propose a wearable input method by touching a face. In the technology of wearable computing, the input space is limited due to the small factors of the devices. In the previous research, we showed that the face as the input space by rubbing the various parts of the face surface by hand with the eyewear device. However, this method only allows simple commands such as a switch. In this research, we propose a method that enables more diverse and continuous input using the eyewear device.

Machine learning assisted genome-wide analysis of chromatin accessibility in ovarian cancer

Hidenori Machino (Trainee,Division of Molecular Modification and Cancer Biology,National Cancer Center Reseach Institute)

We will predict the genome-wide profile of transcription factor binding sites by applying machine learning methods to ATAC-seq data of multi-stage high-grade serous ovarian cancer model samples. Then, we will perform ChIP-seq targeting identified transcription factors to evaluate prediction accuracy of each machine learning method and explore potential therapeutic targets for high-grade serous ovarian cancer. In addition, the genome-wide chromatin status data of fallopian tube secretory epithelial cells obtained by this project is to be utilized for machine learning algorism to predict cell-of-origin of ovarian cancer.

Deep learning predictions of music that give moved brain state

Kazuma Mori (Postdoctoral Researcher,Center for Information and Neural Networks,National Institute of Information and Communications Technology)

I has found that the two kinds of moved states (goose bumps, tears) caused by music gave brain activities related to rewards such as food and drug. In this proposal, I establish a method to select music that induce moved states using deep learning, and measure fMRI brain activity when moved states are induced by unknown music. Once this research is completed, it is expected to obtaining unique knowledge of the neural basis of moved states. Further, this research also may contribute to the development of a new music selection service and music composition method based on the brain science.

Structuration Technique on Argument Mining and Its Application for Large-Scale Consensus Building

Gaku Morio (Graduate Student(Doctor's course),Institute of Engineering,Tokyo University of Agriculture and Technology)

This research proposes a novel scheme for Argument Mining, annotates data and provides a novel deep learning model to improve performances. Also, expertise estimation methodology using the created corpus will be developed.

Linguistically-oriented Dataset Creation for Recognizing Textual Entailment

Hitomi Yanaka (Research Scientist,Center for Advanced Intelligence Project,RIKEN)

Recognizing Textual Entailment (RTE) is a challenging natural language processing task that aims to judge whether one sentence logically entails or contradicts another. Previous studies raise the issue of a lack of datasets for evaluating whether or not RTE systems capture various linguistic phenomena in texts. This study aims to provide a new dataset for evaluating RTE systems by focusing on important linguistic phenomena as studied in formal semantics and pragmatics.

Investigation of Stochastic Regularization for Object Recognition

Yoshihiro Yamada (Ph.D Student,Graduate School of Engineering,Osaka Prefecture University)

The problem of classifying images into categories such as "mountain", "chair", and "airplane" is called general object recognition. General object recognition plays a major role as a fundamental technology, which contributes to the applied technology with great social impact (e.g., automatic driving), and the expectations are high. We achieved the world's highest recognition accuracy on general object recognition by a method of stochastically perturbating the learning process for recognition. On the other hand, this method defies the conventional common sense, and has not been completely explained. Therefore, we aim to experimentally analyze the mechanisms of our method to promote theoretical interpretation.

Development of frameworks for subgraph enumeration from degenerate graphs

Kunihiro Wasa (Assistant Professor by Special Appointment,Principles of Informatics Research Division,National Institute of Informatics)

By improvement of computational performance, we can easily collect huge amounts of semi-structured data, called graphs, regardless of the filed. However, the enormousness makes us difficult to extract useful knowledge from the data. In this study, we set a goal to develop substructure enumeration algorithms that are fundamental techniques for such extraction. In particular, we focus on a parameter of the sparseness of a graph, called the degeneracy.