| 河村 奈々 | M, 1回目発表 | 計算システムズ生物学 | 金谷 重彦, 松本 健一 |
|
title:Unsupervised segmentation of coronary arteries of CTA images using deep learning
abstract: Coronary artery disease is the most common cause of death in the world. For early detection of this coronary artery disease, examination using a CTA image that can contrast blood vessels and identify the stenotic site that is the cause is effective. Exams using CTA images are increasing in number year by year, and it is desired to increase the diagnosis time in order to examine many patients. As a process capable of shortening the diagnosis time, there is a process of segmenting a coronary artery from a CTA image. This operation is performed manually or semi-automatically using a software, and is not fully automated, and is one of the operations that spend doctor's time. Therefore, it is desired to fully automate this work, and many studies are actually being conducted. In particular, many studies using deep learning have been conducted recently. The problem in these studies is little learning data. When performing segmentation by supervised learning, it is necessary to prepare a labeled data set, but it takes a lot of time to create that. In order to solve this problem, this study is to perform coronary artery segmentation with unsupervised learning. language of the presentation: Japanese | |||
| 滝下 雄太 | M, 1回目発表 | コンピューティング・アーキテクチャ | 中島 康彦, 井上 美智子, 中田 尚, Tran Thi Hong, 張 任遠 |
|
title:Neuromorphic Computing using xbar synapse array memristor
abstract:In recent years, the number of IoT devices has increased and is expected to increase in the future. However, the power consumption of the current computer system is large, and the size and cost of the device are large. Therefore, low power consumption is assumed for use at the edge by using the electrical characteristics of the memristor as a synapse. A simple neural network. language of the presentation: Japanese | |||
| 田中 智基 | 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, 笠原 正治, 諏訪 博彦, 藤本 まなと, 松田 裕貴 |
|
title: Evacuation center determination method considering features of victims and environments of disaster areas.
abstract:In disaster situations, appropriate and effective evacuation methods are essential to keep victims safe. Most traditional evacuation methods lead victims to the nearest shelters and consider that all victims have the same mobility. However, in actual disaster situations, vulnerable people like children, elderly people, pregnant mothers and handicapped people will exist. Therefore, considering mobility of victims and road environments including stairs and slopes is required for evacuation guidance. In addition, due to fires and traffic jams after disasters, road environments change as time goes by. This means dynamic changes of road environments in disaster areas are also required to be considered. In this researche, we propose an evacuation center determination method considering features of victims and environments of disaster areas. language of the presentation: Japanese 発表概要: *** この部分を発表概要に *** | |||
| 上田 葵 | M, 1回目発表 | ディペンダブルシステム学 | 井上 美智子, 浦岡 行治(MS), 大下 福仁, 新谷 道広 |
|
title:Measurement and modeling of aging degradation for SiC MOSFET
abstract:Silicon carbide (SiC) is one of the most promising materials to realize high-frequency switching power converters. Compared to silicon (Si), which has been used as the material of power devices, SiC has superior properties, such as high breakdown voltage and high electron mobility. However, the long-term reliability issue of SiC device is of a concern owing to influence of oxide trap charging, thus it is crucial to predict the degradation of circuit performance at the early stage of design. Our research aims to reproduce the performance degradation of power converter using SiC MOSFET by circuit simulator. For this purpose, we will model the BTI degradation mechanism based on actual measurement using SiC MOSFET, and evaluate the model comparing among simulations and measurements on power converter. In this presentation, the experimental results on commercial SiC MOSFET show the degradation of the threshold voltage. language of the presentation: Japanese 発表題目:SiC MOSFETにおける経年劣化の測定とモデル化 発表概要:電力の高効率利用に対する需要の高まりから，SiC（炭化ケイ素）MOSFETを用いた電力変換回路が期待されている．従来から用いられているSiと比較して， SiCは，絶縁破壊電界強度，飽和電子速度，熱伝導等の特性が優れている．一方で，SiC MOSFETは界面の欠陥に起因する長期信頼性課題（特に，Bias Temperature Instability，BTI）が指摘されており，設計段階で回路性能の劣化を見積もることが強く求められている．本研究では，回路シミュレーションを用いてSiC MOSFETを用いた電力変換回路の性能劣化を見積もり可能とすることを目的とする．具体的には．BTIに起因するしきい値電圧の変動を観測し， これをモデル化することで回路シミュレーションモデルに組み込み，最終的に電力変換回路を用いた評価を行う．本発表では，市販SiC MOSFETに対するしきい値電圧変動の実測結果を報告する． | |||
| 酒井 翠 | M, 1回目発表 | 数理情報学 | 池田 和司, 佐藤 嘉伸, 吉本 潤一郎, 久保 孝富, 福嶋 誠 |
|
title: An exploratory study on morphological features of brain MRI relevant to subtypes of psychiatric depression
abstract: Major depressive disorder, also referred to as depression, is a disease that causes social loss next to cancer. However, disease state in depression remains unclear compared with cancer. Most of the research is still continuing on how biological information can be used to enhance diagnosis, treatment and prognosis. The heterogeneity of depression makes it difficult to find biomarkers for depression. Depression can be broken into categories depending on the symptoms. Typical depression is known as "melancholic depression". As symptoms, it is often accompanied by having negative feelings such as anger and guilt, long-term depression of mood from various causes. n the other hand, new subtypes of depressive episode, referred to as "atypical depression", are increasing in recent years, whose symptoms are different from typical depression. 発表題目:うつ病サブタイプの差異に関連した脳MRIの形態的特徴に関する探索的研究 発表概要: うつ病性障害（うつ病）は，がんに次ぐ社会的損失の原因となっている疾患である． しかしながら、がんと比べるとうつ病の病態は未解明な部分が多く，診断や創薬に活用できるバイオマーカーはいまだ確立されていない． その困難さの原因の一つとして，うつ病の異質性が高さがしばしば指摘されている． 症状的に見ると，うつ病はいくつかの病型に分類することができる． 代表的な病型は「メランコリー型」と呼ばれるものであり，持続的な抑うつ気分や意欲低下，食欲不振などによってに特徴づけられる． 一方で，近年では，気分反応性や過食・過眠といったメランコリー型では説明がつかない「非定型」に分類される症例も増えてきている． これらの病型の違いは，脳の構造的な回路異常の差異によって説明付けられるだろうか。 その疑問に答えるために，我々は，広島大学病院およびその関連医療機関を受診した発症早期のうつ病患者148名の脳構造MRI画像を撮像し， その形態的特徴量をメランコリー型患者群と非メランコリー型患者群の間で比較した． 本報告では，その予備結果とそれによって明らかになった解析上の問題点について報告する． language of the presentation: Japanese | |||
| 佐々木 文博 | M, 1回目発表 | 数理情報学 | 池田 和司, 林 優一, 吉本 潤一郎, 福嶋 誠 |
|
title: Developing the model to estimate the probability of the bankruptcy of Japanese enterprises under the assumption of the Wiener process
abstruct: Bankruptcy of the enterprises in Japan is still an economical problem. Generally, there are two categories of the enterprises; Private enterprises, Listed enterprises. However, the private enterprises are not publicly listed. Thus, we cannot estimate a probability of bankruptcy of these enterprises from the stock price change. However, we can use Balance Sheets(BS), Cash Flow Statements(CFS), Profit and Loss Statements(PLS), etc. of the enterprises as the financial index for estimating the probability of the bankruptcy. There is some methodology for estimating the probability; Logit analysis, Problt analysis, Linear regression analysis, etc. However, the previous research imply the probability of the bankruptcy follow the Wiener process. Thus, we think we can inference the probability of the bankruptcy on the premise of the Wiener process. In this research, We focus to find the model for inferencing the probability of the bankruptcy of Japanese enterprises under the assumption of that the enterprises' surplus fluctuation follow the Wiener process. We aim to use the dataset for constructing this model from "Japan Company Handbook". language of the presentation: Japanese | |||
| 設樂 一碩 | M, 1回目発表 | 知能コミュニケーション | 中村 哲, 松本 裕治, 田中 宏季 |
|
title:Automated Cognitive Behavior Therapy by using Embodied Conversational Agent
abstract:Cognitive Behavior Therapy (CBT) is an interactive treatment used for depression and anxiety disorders. CBT has an educational aspect and has the advantage of helping prevent recurrences. In the previous studies, a text dialogue service that partially applied CBT was developed, and its effectiveness has been proven. However, in terms of omission of treatment procedures and text-only interaction, it is difficult to say that the effects of face-to-face treatment are sufficiently brought out. Therefore, in this study, we are going to analyze behavioral characteristic information from interactive data during human-to-human treatment and create a conversation function based on medical evidence. After that, we are going to implement it in an Embodied Conversational Agent (ECA) and attempt to automate CBT considering verbal/non-verbal information. language of the presentation: Japanese | |||
| 河合 将隆 | M, 1回目発表 | サイバーレジリエンス構成学 | 門林 雄基, 中村 哲, 宮本 大輔（東京大学）, 妙中 雄三 |
|
title: Study on giving explanations in log monitoring using machine learning
abstract: While it is said that there is a shortage of cybersecurity personnel, companies are demanding “people who can monitor and analyze logs”. For this reason, a log monitoring technique using machine learning has been studied. However, the conventional technique does not explain the basis for judgment on the classification result, and lacks reliability on actual operation. Further, since only the classification result is output, there is a problem that there is no detailed information about the result and it is difficult to examine a countermeasure method. Therefore, in this study, in log monitoring using machine learning, we conduct research for the purpose of giving explanation to the classification result. language of the presentation: Japanese 発表題目: 機械学習を使用したログ監視における説明性の付与に関する研究 発表概要: サイバーセキュリティ人材が不足していくと言われる中、企業では「ログを監視・分析できる人」が求められている。そのために、機械学習を使用したログ監視手法が研究されているが、従来の手法では、分類結果に対する判断根拠の説明がなく、セキュリティシステムとして実運用を考えた時に信頼性に欠ける。また、分類結果のみを出力するため、結果に関して詳細な情報がなく、対策方法の検討が難しいといった問題がある。そこで本研究では、機械学習を使用したログ監視において、分類結果への説明性の付与を目的として研究を行う。 | |||
| 石川 武典 | M, 1回目発表 | 光メディアインタフェース | 向川 康博, 清川 清, 舩冨 卓哉, 田中 賢一郎, 久保 尋之 |
|
title: Distance estimation of two-layer structure object using multiple modulation frequencies in time-of-flight measurement
abstract:In recent years, Time-of-Flight(ToF) cameras have been used in various fields.However, this camera cannot measure the correct distance in a two-layer scene where a translucent object and an opaque object overlap. For example, products packed in objects and films over glass. Therefore, we propose a method for estimating the correct distance both translucent and opaque objects by using multiple modulation frequencies. In this time, the effectiveness of the proposed method was confirmed through simulation experiments. language of the presentation: Japanese | |||
| 小林 誠人 | M, 1回目発表 | サイバネティクス・リアリティ工学 | 清川 清, 小笠原 司, 酒田 信親, 磯山 直也 |
|
title: A Flying Sensation Display by Using a Jet Pool and Underwater VR
abstract:In recent years, VR technology has made it possible to experience flying in the sky with VR. However, the present flight sensation presentation system cannot present a feeling of floating in the air. On the other hand, by experiencing VR underwater, it is considered possible to present a sense of flight with floating feeling in the air utilizing buoyancy. Therefore, in this study, we will present a flying sensation with floating feeling in underwater VR. In order to present a sense of speed underwater this time, we implemented an HMD that can be used underwater, designed a system, and created presentation content. language of the presentation: Japanese | |||
| 髙橋 秀明 | M, 1回目発表 | サイバーレジリエンス構成学 | 門林 雄基, 中島 康彦, 笠原 正治, 妙中 雄三 |
|
title: Examination of security mechanism in NFV MANO
abstract: 5G is attracting attention as a communication infrastructure for society5.0. The NFV (network virtualization) used in the core network has been actively developed in recent years. However, since they are mainly implemented, it has been found that there are almost no security mechanisms. In this presentation, I will explain the outline of NFV, the contents of the investigation, and future issues. language of the presentation:Japanese 発表題目: NFV MANOにおけるセキュリティ機構の検討 発表概要:Society5.0の通信基盤として注目される５G。 そのコアネットワークにおいて用いられるNFV（ネットワーク機能の仮想化）であり、その開発が近年活発になっている。 しかしながら、それらは、実装を主に進められているため、セキュリティ機構がほとんど存在しないことが分かった。 本発表では、NFVの概要からその調査内容と今後の課題までを説明する。 | |||
| 安本 玄樹 | M, 1回目発表 | 知能コミュニケーション | 中村 哲, 松本 裕治, 須藤 克仁 |
|
title: Generating Context Sentences toward Context-Aware Machine Translation
abstract: Recent developments in machine translation have been remarkable. There are reports that sentence-level machine translation has achieved the same level as human translation. However, evaluation of document-level machine translation is still inferior to human translation. The purpose of our research is to propose a document-level machine translation model. In previous researches, there are many models that improve the accuracy of translation using information of previous k sentences. However, these models cannot take into account long or many sentences because of lack of memory or difficulty to control attention mechanism. In this research, we try to make "context sentences" to solve these problems. language of the presentation: Japanese | |||
| 龍田 侑弥 | M, 1回目発表 | ロボティクス | 小笠原 司, 杉本 謙二, 高松 淳, Gustavo Garcia |
|
title: In-hand manipulation by using universal gripper
abstract: Manipulator needs more universality of grasping because of increasing types of objects. The soft gripper whose grasping point is soft can decide grasp point with the object flexibly in comparison with solid grippers. However, it is difficult for soft grippers to do in-hand manipulation because soft manipulators consider the transformation of the elastomer. Therefore in these years, there are some studies to acquire elastomer's transformation from camera information. On the other hand, it was difficult to acquire elastomer's transformation because there were filling in the elastomer in universal gripper using the jamming transitions. Now we succeed in the acquisition of the elastomer's transformation by using acryl filling and the fluid of the same refractive index. In this presentation, I explain my study plan to achieve In-hand manipulation using Universal gripper. I am going to perform In-hand manipulation by the pressure control of the internal fluid and the estimate of the torque of the grasped object. I will focus on reversing the object task as a concrete task. language of the presentation: Japanese | |||
| 中西 直樹 | 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: Understanding the musculoskeletal structures of lower extremity from a radiograph using deep learning
abstract: Decomposition of musculoskeletal structures of lower extremity from medical images is useful for quantitatively understanding the process of muscle atrophy caused by disease and the recovery process during the rehabilitation. In this study, we focus on decomposition of individual musculoskeletal structures from single radiograph that can be acquired with high spatial resolution and low radiation dose. Then, we treat the extraction problem in real radiograph as the image translation problem from a real radiograph to multiple digitally reconstructed radiographs generated using CT image and the 3D mask of individual muscles which was obtained by a previously proposed automatic segmentation method. Since registration of real and synthetic radiographs is a challenging problem especially for the muscle structures, we introduce an image translation method based on unpaired training data set using CycleGAN. In this study, simulation and real image experiments were conducted using CT images and real radiographs of 475 cases. language of the presentation: Japanese | |||
| REHMAN IFRAZ | M, 1回目発表 | ソフトウェア工学 | 松本 健一, 笠原 正治, 石尾 隆, 畑 秀明, Raula G. Kula |
|
title: An Analysis of Newbies Contribution to Projects in GitHub
abstract: The success of Open Source Software (OSS) heavily relies on the existence of a project’s ability to attract and sustain newcomers (i.e., newbies) contributions. Related work show that newbies’ onboard and growth patterns (attraction) relies on the maintenance of some specific factors of OSS projects. Yet, the problem is that we do not yet understand what kinds of projects becomes the source of attraction for newbies. In this study, we would like to understand the different kinds of projects that attracted newbie, with the goal of helping to sustain newbie contributions. Our approach is to empirically study the projects of first time committers to GitHub. In detail, first from 11,000 newbies, we mine the histories of their repositories to understand the rate of contributions (i.e., speed) over time. Second, we then qualitatively, take a sample of these contributors to understand the types of projects they contribute. As a result, we classify the projects in three category, i.e., toy or not toy or fork, in the respect of their rate of contributions. Comparatively, we found significant differences in the different types of repositories based on their speeds. language of the presentation: English | |||
| 溝江 俊太郎 | 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: Automated analysis of 3D dynamics of foot and ankle joints using robust CT segmentation and multi-rigid 2D-3D
registration.
abstract: Three-dimensional dynamic analysis of bone is important in the diagnosis and treatment of orthopedics.As an analysis method, 2D-3D registration using CT and X-ray moving images has been proposed.However, for the analysis of complex joints composed of many bones such as tarsal bones, it was necessary to manually perform CT segmentation of individual bones and initial positioning for registration.The clinical routine application for a large number of patients was not realistic.Therefore, in this study, 3D position estimation was performed by CT segmentation and landmark extraction using CNN and landmark extraction from two-way X-ray moving images for full automatic analysis.Using this result as the initial position in 2D-3D registration based on higher-precision luminance values, we aim to develop a system that automatically analyzes the dynamics of the foot and ankle joint. language of the presentation: Japanese | |||
| 武田 敏季 | M, 1回目発表 | 知能システム制御 | 杉本 謙二, 池田 和司, 小林 泰介 |
|
title: Towards Disentangled Latent State Space for Distributed Control
abstract: In recent years, the systems handled by control engineering have become remarkably larger and more complex, but they require intractable system identification. Machine learning is one of the promising approaches to system identification problems, hence it has actively been studied to extract latent state space hidden in data, which can reconstruct system characteristics. However, because such extraction is performed in the unsupervised manner, the latent state space is not necessarily able to apply to the state-of-the-art control methods. This study, therefore, develops the way to shape the latent state space suitable for control. Specifically, we focus on the disentangled (i.e., independent) latent state space, which would make distributed control possible. To this end, a variational autoencoder (VAE) with Laplace distribution, which is highly non-Gaussian and used in classical independent component analysis, is proposed. In a benchmark problem, the proposed VAE gained high independency while holding the input information. language of the presentation: Japanese | |||
| ILBOUDO WENDYAM ERIC LIONEL | M, 1回目発表 | 知能システム制御 | 杉本 謙二, 池田 和司, 小林 泰介 |
|
title: T-Adam, a new robust stochastic gradient optimization method.
abstract: Machine learning algorithms aim to find patterns from observations, which may include some noise. To perform well even with such noise, we expect them to be able to detect outliers and discard them when needed. We therefore propose a new stochastic gradient optimization method, whose robustness is directly built in the algorithm, using the robust student-t distribution as its core idea. Adam, the popular optimization method, is modified with our method. The obtained optimizer, T-Adam, effectively outperforms Adam in terms of robustness against noise on simple tasks, such as a linear regression problem, and also on more complex tasks, such as a reinforcement learning problem. language of the presentation: English | |||