| 福田 修之 | M, 1回目発表 | ユビキタスコンピューティングシステム | 安本 慶一, 池田 和司, 荒川 豊（客員）, 松田 裕貴 |
|
title: Analysis of relationship between sleep status and DAMS questionnaire of office workers
abstract: In recent years, researches focusing on physiological states of office workers have been actively conducted for the purpose of measuring intellectual productivity and stress state. The purpose of this study is to clarify the relationship between sleep conditions obtained from wearable devices and occupational health indicators. In this report, we focus on the results of the DAMS questionnaire, which is an index for investigating the ups and downs of mood, and report the results of machine learning analysis. language of the presentation:Japanese | |||
| 山口 泰弘 | M, 1回目発表 | 自然言語処理学 | 松本 裕治, 中村 哲, 新保 仁, 進藤 裕之 |
|
title: Polymer-Solvent Relation Extraction from Scientific Papers
abstract: A polymer database plays important role in the development of material informatics. The database is created by extracting information from scientific papers. The goal of this study is extracting relations between polymers and solvents from scientific papers automatically. In this report, we build named entity recognition model and relation extraction model for polymers and solvents. We evaluate the performance of these models and analyze its predictions. language of the presentation: Japanese | |||
| 有元 遼 | M, 1回目発表 | 知能システム制御 | 杉本 謙二, 岡田 実, 小蔵 正輝 |
|
title: State Estimation with Gain Switching Observer against Signal Losses
abstract: Networked control systems, where communication signals are transmitted through a shared network, have advantages of design of flexible system but there are problems of signal losses. To deal with performance degradation caused by signal losses, we design an observer and consider the situation by measuring the output of the plant. We switch the observer gain in synchronization with the scheduling of communication and improve control performance. In this report, we verify the idea of our proposed method and suggest the direction of the future experiment. language of the presentation: Japanese | |||
| 稲葉 光彦 | M, 1回目発表 | 数理情報学（計算神経科学） | 池田 和司, 作村 諭一(BS), 川鍋 元明（客員）, 森本 淳（客員）, 福嶋 誠 |
|
title:Decoding one-dimensional arm movements from EEG under realistic feedback training
abstract:Brain-machine interface (BMI) is a technology of operating external devices by using changes in brain activity. In non-laboratory environments, electroencephalogram (EEG) and near-infrared spectroscopy (NIRS) are the main measurement methods for BMI. Non-invasive BMIs with EEG and NIRS usually employ mental states that do not match intended actions in order to achieve high accuracy. Such lack of “intuitiveness” in BMI causes worse operability and longer training time. As a first challenge towards more intuitive BMI, we aim to decode users’ intended directions of single arm movements. In this talk, we will explain a planned EEG experiment with realistic visual feedback for “intuitive” BMI training. language of the presentation:Japanese | |||
| 佐藤 太清 | M, 1回目発表 | 自然言語処理学 | 松本 裕治, 中村 哲, 新保 仁, 進藤 裕之 |
|
title: Evaluating and Generating Japanese poem Tanka Automatically
abstract: The goal of this study is generating "good" Japanese poem tanka automatically. Generating poem is challenging task but many study are trying to create poem like "human made". For generating "good" Tanka, we made a big Tanka corpus, each Tanka evaluated by human. We also make automatic Tanka evaluating system in step of making generating "good" Tanka. The evaluating system support improving tanka. In this presentation, we show the outline of this study and the first step, how to extract feature of Tanka. language of the presentation: Japanese | |||
| 上里 勇希 | M, 1回目発表 | 知能コミュニケーション | 中村 哲, 松本 裕治, Sakriani Sakti |
|
title: Semi-supervised spoken style learning with Speech Chain Framework
abstract: A machine speech chain was proposed to mimic human auditory feedback from the speaker’s mouth to her ears that was implemented as a closed-loop chain mechanism with both automatic speech recognition (ASR) and text-to-speech (TTS) models. The framework can achieve semi-supervised learning by enabling each component to assist each other when they receive unpaired data, which then allows them to infer the missing pair and optimize the models with reconstruction loss. But, the current machine speech chain only deals with either synthetic speech or natural read speech with a monotonous style. However, in reality, humans speak with various forms and expressions. In this research, we propose to extend the capability of the machine speech chain to learn different spoken styles in a semi-supervised manner. Specifically, we will investigate how to construct style embedding with deep learning. By leveraging this framework, we may empower TTS to be style-aware, and ASR to be robust for the prosodic difference. In this talk, I outline the background, the proposed model framework, and the future research plan. language of the presentation: Japanese | |||
| 櫻井 和貴 | M, 1回目発表 | サイバーレジリエンス構成学 | 門林 雄基, 笠原 正治, 岡田 実, 妙中 雄三 |
|
title: Development of the anti-jamming and anti-tamper defense mechanism for automatic identification system
abstract: As an island country, Japan imports most of its resources by ship. Therefore, ensuring smooth maritime traffic is an important issue in the national strategy. Due to the characteristics of maritime traffic, there are some limitations to install signs and signals, unlike land roads. Therefore, in many cases, equipment mounted on each ship takes on a role in preventing ship collision. Among them, AIS (Automatic Identification System) is in charge of exchanging each other's navigation information such as position, course, and speed. However, AIS has no defensive measures from the viewpoint of network security because of its specifications and can be easily intercepted and tampered with. If navigation information is not exchanged properly, smooth marine traffic may be hindered, and in the worst case, it may lead to ship collisions. I will explain the development of anti-jamming and falsification prevention systems and future research plans. language of the presentation: Japanese 発表題目: 船舶自動識別装置に対する通信妨害及び改ざんへの耐性を備えた防御機構の開発に関する研究 発表概要: 島国である我が国は資源の大部分を船舶による輸入に頼る現状があり、円滑な海上交通の確保は国家戦略上の重要課題である。海上交通はその特性から陸上の道路のように標識や信号を設置することには限界がある。そのため衝突防止の仕組みを各船舶に搭載された機器に頼る部分が多くを占める。その中で位置、針路、速力、といった船舶の航行情報を相互交換するために用いられているのが船舶自動識別装置である。しかしながら、船舶自動識別装置は仕様上、ネットワークセキュリティの観点からの防御策は皆無であり容易に通信妨害や航行情報の改ざんが可能である。航行情報が適切にやり取りされない場合、円滑な海上交通に支障をきたし、最悪の場合には船舶の衝突につながる可能性も考えられる。そこで本研究では船舶自動識別装置に対する通信妨害及び改ざんへの耐性を備えた防御機構の開発及び今後の研究実施計画についてのべる。 | |||
| USAWALERTKAMOL BUNYAPON | M, 1回目発表 | ロボティクス | 小笠原 司, 清川 清, 高松 淳, Gustavo Garcia |
|
title : Blind spot moving obstacle detection for mobile robot using sdound-reflection sensor.
abstract: Local path planning is the method for collision avoidance function of mobile robot by using camera-sensors to collected the informations of prediction of obstacle in future step and direction of obstacle, but the camera-sensors is not be able to detect any moving obstacle from the blind spot without additional function of robot. This research will propose the use of sound-reflection sensors in mobile robot to improve efficiency of local path planning by using the advantage of sound wave that can travel in non-straight direction to detecting the moving obstacles behind the blind spot of environments. language of the presentation : English | |||
| 森田 道成 | M, 1回目発表 | 生体医用画像 | 佐藤 嘉伸, 加藤 博一, 大竹 義人, スーフィー マーゼン, 日朝 祐太 |
|
title: 2d-3d reconstruction of skeletal micro-anatonmy from lower limb x-ray images using deep learning
abstract: CT image describe 3D information. Thus, it is widely used for diagnostic imaging and preoperative planning. However, CT scan involves relatively high medical cost and radiation exposure. On the other hand, single radiograph can be acquired with lower radiation exposure and lower cost than CT scan, while it can obtain only 2D information. In this work, we aim at 2D-3D reconstruction of CT images from single radiograph. In the proposed method, firstly, we synthesize a digitally reconstructed radiographs(DRR) of the target bones from single radiograph using GAN. Then, we reconstruct CT images from the synthesized DRR using deep learning. In this paper, we discuss the preliminary result of 2D-3D reconstruction of CT image from DRR of pelvic area. language of the presentation: Japanese | |||
| 萩岡 宣旭 | M, 1回目発表 | 生体医用画像 | 佐藤 嘉伸, 小笠原 司, 大竹 義人, スーフィー マーゼン, 日朝 祐太 |
|
title: Biomechanical Simulation Model with Muscle Fibers Derived using Muscle Fiber Tractography of High-Resolution Images
abstract: Biomechanical simulation models are built from linear muscle fibers as actuators. However, the fibers are derived without considering the internal structure characteristics, such as fiber orientation, and it is expected that the accuracy of the simulation would be improved by reflecting these characteristics. We propose to use a fiber tractography approach to capture the internal muscle structure. We create a simulation model that reflects the muscle structure using the muscle fiber tractography and report the results of the study. language of the presentation: Japanese | |||
| 本田 修平 | M, 1回目発表 | 生体医用画像 | 佐藤 嘉伸, 向川 康博, 大竹 義人, スーフィー マーゼン, 日朝 祐太 |
|
title: Analysis of disease classification and musculoskeletal anatomy using medical images and radiology reports in a large-scale medical image database
abstract: Recently, the environment for the analysis of large databases, such as the large-scale medical image database, have been constructed. Currently, the development of a system that learns medical images and finding diagnostic sentences from reports, and estimates the finding sentences for new images, is being investigated. However, little research has been done to extract clinically useful information from both medical images and the findings. In this study, CT images of about 10,000 cases, radiology reports, and patient attribute information such as age and gender accumulated in the medical big data cloud platform built at the National Institute of Informatics (NII) are used. The aim is to analyze and clarify the relationship between the disease classifications extracted by a natural language processing approach from the radiological interpretation report and the parameters related to the anatomical shape of the musculoskeletal system obtained by means of segmentation by deep learning from CT images. language of the presentation: Japanese | |||