| 湯川 大雅 | M, 1回目発表 | 情報基盤システム学 | 藤川 和利 | 門林 雄基 | 新井 イスマイル | ||
|
title: Intrusion Detection System for Automotive Networks using Two-Stage Machine Learning Model
abstract: In recent years, electronic controlled vehicles, including automated and connected cars, have been popular. A number of electronic control units (ECUs) are installed in vehicles that are controlled electronically. For their interconnection, controller area network (CAN) is commonly used. However, since CAN has no security feature to protect communications from attacks, vehicles and the users can be damaged through attacks such as DoS and injection of malicious messages. Intrusion detection has been studied as one of the methods to defend against such attacks. For safety reasons, real-time and accuracy are important for intrusion detection in automotive networks. In addition to this, the processing must be done with limited resources. In this presentation, I will introduce some studies of intrusion detection in automotive networks and an intrusion detection model, the Two-Stage Deep Learning model (TSDL), that combines two deep learning models. Finally, I propose to use a model based on the TSDL model for intrusion detection in-vehicle networks. The proposed model select the two models of the TSDL model dynamically and is expected to improve both real-time and detection performance in automotive networks. language of the presentation: Japanese | |||||||
| 澤田 篤志 | 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 | TRAN THI HONG | 張 任遠 | |
|
title: Development of neuromorphic system using mem capacitor
abstract: In recent years, the performance of machine learning has improved, but along with this, the increase in power consumption and computational complexity has become a problem. On the other hand, the speed of miniaturization of semiconductor manufacturing technology is slowing down, and in order to solve these problems, it is necessary to switch to a new architecture. Neuromorphic systems are attracting attention as an information technology platform for solving the above problems. The neuromorphic system is a system that mimics the structure of nerve cells in the brain. Conventionally, CMOS was used for the element of the network, but in this research, we propose a circuit that uses a MEM capacitor for the element and a charge pump circuit for the spike operation. The circuit can be integrated by using a MEM capacitor whose resistance value changes depending on the time and value of voltage application, and the charge pump circuit is used to enable spike operation, aiming to realize low power consumption. language of the presentation: Japanese 発表題目: メムキャパシタを用いたニューロモーフィックシステムの開発 発表概要: 近年, 機械学習の高性能化が進んだが, それに伴い消費電力や計算量の増加が問題となっている。一方, 半導体製造技術微細化のスピードは低下しており, これらの問題を解決するためには, 新しいアーキテクチャへの転換が必要である。ニューロモーフィック・システムは, 上記の問題を解決するための情報技術基盤として注目されている。 ニューロモーフィックシステムは脳の神経細胞の構造を模倣したシステムである。 従来ではネットワークの素子にCMOSが用いられていたが, 本研究では, 素子にメムキャパシタを用い, スパイク動作にチャージポンプ回路を用いた回路を提案する。電圧印加の時間や値によって抵抗値が変化するメムキャパシタを用いることで回路の集積化が可能となり, チャージポンプ回路を利用してスパイク動作を可能にすることでを低消費電力化の実現を目指す。 | |||||||
| 綿貫 零真 | M, 1回目発表 | 知能システム制御 | 杉本 謙二 | 小笠原 司 | 小林 泰介 | ||
|
title: Active-sensing-used Reinforcement Learning on Partially Observable Human State
abstract: Reinforcement learning has been applied to the acquisition of complex robot behaviors.However, it assumes that an agent is on fully observable of environment’s state, while many actual environments are regarded as partially observable Markov decision processes (POMDP). In particular, human state cannot be observed perfectly using robot’s sensors. As a solution to POMDP, the use of time-series data and active sensing, in which agents actively improve their own observation information by selecting what/where/when/how is observed, are effective. In this presentation, examples show that reinforcement learning fails in POMDP and suitable observation mitigates the difficulty. language of the presentation: Japanese | |||||||
| 伊東 尚輝 | M, 1回目発表 | 生体医用画像 | 佐藤 嘉伸 | 金谷 重彦 | 大竹 義人 | Soufi Mazen | 上村 圭亮 |
|
Title: Automated muscle segmentation in CT images using few training datasets : Preliminary experiments
Abstract: Patient-specific functional analysis of the musculature is necessary in several applications. In our previous studies, we have achieved accurate muscle segmentation at the region around the hip joint. Therefore, the final goal of this study is to perform accurate whole-body muscle segmentation. In addition, since it is a difficult task to create whole-body ground-truth (teaching) data, we aim to achieve muscle segmentation by using smaller amount of ground-truth data. In order to achieve this goal, we investigated the impact of a typical deep learning-based image segmentation model, U-net, on the accuracy of segmentation by analyzing its performance under various hyper-parameter settings Language of the presentation: Japanese | |||||||
| 豊田 真行 | M, 1回目発表 | ロボティクス | 小笠原 司 | 加藤 博一 | 高松 淳 | 趙 崇貴 | |
|
title: Elucidating essential elements for an end-effector of a touch-care robot
abstract: Affective touch enables to relieve stress and pain, and thus has a positive effect on dementia and autism relief. If robots that can perform affective touch would be achieved, the robots would support the patients who cannot receive care from therapists. Though the softness and the warmness like a human hand are required for an end-effector for touch, other requirements are still unclear. In this study, we elucidate essential elements of designing the end-effector, which can provide affective touch equivalent to or more than human touch. The previous research pointed out the relationship between the comfort of affective touch and the contact area. Focusing on the area and distribution of the contact surface on touch, we aim to verify whether human feeling changes when we change these elements. We consider the design of the end-effector to fit various body parts. language of the presentation: Japanese | |||||||
| MISHRA VIPUL | M, 1回目発表 | ソーシャル・コンピューティング | 荒牧 英治 | 渡辺 太郎 | 若宮 翔子 | ||
|
title: Are Metal Fans Angrier than Hip-Hop Fans?:
Comparative Study of Emotional States of Fan Communities of Different Music Genre
abstract: Humans have a fascinating relation with music, and music forms a part of our identity. Thus, people with shared taste in music should have some shared identity or traits. As far as we know, there is no prominent study relating human emotion and human behavior with music taste at large scale. The goal of our research is to analyze and compare the traits of groups following different styles of music. In order to investigate this, we examine the emotional tone of utterances on online fan communities. We collect comments from Reddit and YouTube and run emotion analysis on them using existing SoTA methods. We expect analysis of the emotional tones of the utterances from different genres to elucidate some aspects of relation of humans with music. language of the presentation: English | |||||||
| 齊藤 晴香 | M, 1回目発表 | 光メディアインタフェース | 向川 康博 | 安本 慶一 | 舩冨 卓哉 | 田中 賢一郎 | |
|
title: Consideration of miniaturization and wide-angle imaging of camera system by small aperture imaging
abstract:For smartening of home appliances and housing equipment, sensing devices that use the behavior recognition of people in the house have being researched. In this study, we make the camera system smaller and its view angle wider by devising the optical system. It is expected that the camera system should be smaller and have wider view angle for the behavior recognition in the house. As the first approach, we use a lens-less method in which the angle of the image is bent with a diffraction grating. By the experiment, it is possible to observe a wider angle of view by obtaining image of the invisible part due to limb darkening. As a second approach, a wide-angle image could be obtained with a plano-convex lens close to the image sensor. language of the presentation: Japanese | |||||||
| TIAN JIAXING | M, 1回目発表 | 数理情報学 | 池田 和司 | 作村 諭一 | 吉本 潤一郎 | 福嶋 誠 | 日永田智絵 |
|
title: An research for depression counseling based on emotional models
abstract: Nowadays, depression is increasing year by year. If there is a method that can simulate the physical feature and the pathological process of depression, it will be helpful to the treatment of depression. Currently, Stephan has proposed a computational model explained the depression based on the theory of predictive coding and activity inference. The model can explain the relationship between fatigue and homeostasis and predict the adjustment process of homeostasis. However, this model is not ideal for the actual physical feature data. This research will collect the actual physical feature data. Based on the Stephan model, we will propose a model that can fit the actual physical feature data. Through this research, we will finally establish a model for effective treatment of depression. language of the presentation: English | |||||||
| POPOV NIKOLAY | M, 1回目発表 | 数理情報学 | 池田 和司☆ | 作村 諭一 | 川鍋 一晃(客員） | 森本 淳(客員） | |
|
Title: Identifying biomarkers of aggressive behaviour with functional and structural MRI features. Abstract: Aggressive behaviour, which is defined by Konrad Lorenz as “the fighting instinct in beast and man which is directed against members of the same species”, on the one hand playing an important role in evolutionary biology, i.e. having adaptive value in many natural environments, on the other hand could take forms of maladaptive behaviour (behaviour which is excessive and inadequate to the circumstances the organizm is put into), which is detrimental not only to the organizm but to the society as a whole in case of homo sapiens. The researchers approach this issue from different prospectives: neurobiological, psychological, ethological, genetical and from the neuroimaging prospective. Still, the accuracy remains low, and the necessity to further explore the issue is looming. Our work is meant to augment the existing features with new structural and functional biomarkers, increase the accuracy of prediction, especially when it's related to identifying proactive (pre-planned)aggressive behaviour (e.g. planned murder, physical abuse). Language of the presentation: Japanese | |||||||
| 宮内 大輝 | M, 1回目発表 | ネットワークシステム学 | 岡田 実 | 笠原 正治 | 東野 武史 | DUONG QUANG THANG | Chen Na |
|
title: Experimental Investigation of Rainfall Nowcast using Zenith Total Delay in GNSS
abstract: The zenith total delay (ZTD) obtained during the positioning process in GNSS varies due to the amount of water vapor in the atmosphere. Although the ztd can be considered as a potential possibility of rainfall, variation of ztd results in false alarm frequently. In this paper, differential time series of ztd is newly introduced to rainfall nowcast in order to improve accuracy. Observation data for five months are postprocessed using precise point positioning algorithm, then false alarm rate is experimentally investigated language of the presentation: Japanese | |||||||