ゼミナール発表

日時: 1月7日(月)3限 (13:30-15:00)


会場: L1

司会: 池田篤俊
田辺 雅人 1251060: M, 1回目発表 ロボティクス
title: Discrimination of accident black spots using tracking data of moving vehicles
abstract:The injury accident of the car is concentrated in the specific place, and the spot is called accident black spot. The place is decided based on statistical data, so it is difficult to find potential accident spot. In this presentation, I will propose the method for discriminating accident black spot using tracking data of moving vehicles and explain the related works.
language of the presentation:Japanese
 
赤本 拓也 1251003: M, 1回目発表 ロボティクス

title: Facial expression translation filter to change facial expression

abstract: Big data of facial expressions is needed to make various facial expressions by Android robot for talking with human.It is useful to generate data of various expressions from data of few expressions.

We make a expression translation filter to change facial expression.In this presentation, I will review two previous researches about the translation filter to change Attribute of motion and the method to make facial expression by android robot.At last I will explain my research plan and progress.

language of the presentation: Japanese

発表題目:表情を多様化させる表情変換フィルタ

発表概要:人とのコミュニケーションを目的としたアンドロイドロボットに様々な表情を表出させるためには,膨大な量の表情データが必要になる.そこで1つの表情データから多様な表情を生成できれば有用であると考えられる.本研究では表情を多様化させる表情変換フィルタを製作する.発表では,データの属性の多様化とアンドロイドロボットの表情表出の先行研究を紹介し,今後の研究の計画と現在の進捗について述べる.

 
中島 往馬 1251076: M, 1回目発表 生命機能計測学
title: Extended techniques of pulse-laser assisted optical tweezers
abstract:Optical tweezers is a technique to trap and manipulate micron size object under microscope by radiation pressure force exerted by a laser beam. This technique is small trapping force. So it is qualified manipulation. We propose that is possible techniques of trap, manipulation, press micron size object into transparent glass substrate, peel from glass substrate and micron size object surmount a steric constraint in the non-contact and non-destructive by strengthening trapping force in a moment using a pulse-laser.
language of the presentation: Japanese
発表題目:パルスアシストを利用した光ピンセットの技術拡張
発表概要;顕微鏡下で光の放射圧を利用して非接触、非破壊で微小物体を捕捉、操作する光ピ ンセット技術がある。この方法は捕捉力が小さく、操作が制限される。我々は、パルスレーザーを用 いて、一瞬の捕捉力を強化することにより、非接触、非破壊のままで捕捉、操作、押し付け、引きはが し、立体障害の突破を可能とする方法を提案する。
 
Munoz Gallardo Nahira 1251129: M, 1回目発表 知能システム制御

Title: Paper introduction: " FTW: Fast Similarity Search under the Time Warping Distance"

Abstract: Electromyographic (EMG) signal processing is commonly used to control prosthetic devices. However, time-variant characteristics of the EMG signal are a major problem that limit the controllability and usability of the protheses. Dynamic time warping (DTW) is a well-known technique for similarity search in time-series data. Although DTW provides scaling along the time axis, its heavy computation cost is prohibitive for real-time control of protheses using EMG signal. In this presentation, I will introduce the FTW algorithm, a fast search method for DTW, which efficiently prunes a significant number of search candidates to reduce the DTW search cost.

Language of the presentation: English

 

会場: L2

司会: Camargo Cruz
小田 将之 1251026: M, 1回目発表 計算メカニズム学
title: Secure Device Pairing Using Accelerometers
abstract: Device pairing is a method for connecting two devices so that they can communicate with each other in a secure way. Entering PIN code is widely used as a general method for device pairing. However, it is troublesome for users and also there are security threats like peeking attack. For this reason, more secure and easier pairing methods have been proposed. Among them is a method for automatically generating a secret key from acceleration data obtained by shaking two devices together. In this presentation, I will review two papers on this topic. In the first paper, a key is generated by time-domain analysis and the second one uses frequency domain analysis. I compare the efficiency and accuracy of those two approaches, and conclude my talk with my future research plan.
language of the presentation: Japanese
発表題目: 加速度センサを用いたセキュアデバイスペアリング
発表概要:現在,デバイスペアリングの一般的な手法として PINコード入力が広く用いられているが,わずらわしく, のぞき見攻撃等の危険がある. より安全で手軽なペアリング手法として, 端末を同時に振って得られる加速度の時系列データから 共通鍵を生成するという方式が提案されている. 本発表ではこのような手法を提案している論文を2つ紹介する. 1つめの論文では時間領域,2つめは周波数領域での特性から 鍵生成を行う.発表ではこれらの2つの手法の鍵生成効率や精度を比較し, 最後に発表者の今後の研究計画について述べる.
 
内垣 聖史 1251017: M, 1回目発表 ソフトウェア工学

title: Early Prediction Model for Fault-Prone Module

abstract: Recently, fault-prone module prediction models are proposed for the efficiency of software testing. However, fault-prone module prediction models are constructed from source code metrics at the time of the coding end. Therefore we cannot discriminate between fault module and non-fault module before completing coding. In this study, we construct the model to predict fault-prone module based on metrics of under coding.

language of the presentation: Japanese

発表題目: Fault-Proneモジュール早期判別モデルの提案

発表概要: 従来,ソフトウェア開発におけるテスト工程の効率化のためにFault-Prone(欠陥が含まれる確率が高い)モジュール判別が注目されている.しかしながら一般的なFault-Proneモジュール判別モデルではコーディング終了時のソースコード特性値に基づいて判別を行うため,コーディング途中の段階では欠陥の有無を判別することはできない.本研究では,コーディング途中の特性値に基づいてFault-Proneモジュールを判別するモデルの構築を試みる.

 
宮本 佳治 1251105: M, 1回目発表 ディペンダブルシステム学
title: Paper introduction : Adaptive Low Shift Power Test Pattern Generator for Logic BIST
abstract: Rapid advances in manufacturing technologies have led to drastic increase in transistor density. Among the many challenges imposed by the technologies, power consumption during test are becoming major concern. I will introduce a paper that present a new low power technique for Built-In Self-Test(BIST).
language of the presentation: Japanese
 
金丸 洋平 1251033: M, 1回目発表 情報基盤システム学
title: Vehicler grouping method for exchanging traffic information
abstract: Exchanging traffic information among nearby vehicles is one of the communication scenarios in the field of intelligent transport systems (ITS). An important question in this scenario is ''How can traffic information be delivered to only one subset of nearby vehicles in the same road traffic situation?'', for example, to only congested traffic in a certain lane. Therefore, constructing a group of relevant vehicle and its method is needed. In this presentation, we describe requirements for such method, and propose a method which utilize a trajectory and orientation. Also, we will show a prototype of our method implemented on Android smartphone.
language of the presentation: Japanese
 

会場: L3

司会: Kevin Duh
志水 友輔 1251052: M, 1回目発表 インタラクティブメディア設計学
title:Appearance control at dynamic light environment using projector-camera system
abstract:Spatial augmented reality not only show annotation, but also can change appearance of real object using projected correction light. In previous study, he used projector-camera system, estimated appearance under white light using projected image and captured image, and made control technique that change target appearance in dynamic environment. However, it is difficult that real object control target appearance in dynamic light environment. So we propose to estimate dynamic ambient light, and to control appearance under more complex environment than previous work. This proposal technique use model of dynamic ambient light, and estimate ambient light and appearance under white light using past two pairs of projected image and captured image.
Language of the presentation:Japanese
発表題目:動的な光環境におけるプロジェクタカメラ系による見かけの制御
発表概要:投影型拡張現実感技術は付加情報の提示だけでなく,補正光重畳投影によって実物体の見えを変調するができる.従来研究では,プロジェクタカメラ系にて動的に変化する状況においても投影画像と撮影画像を用いて白色照明下での見えを推定し,目標の見えとなる制御手法を実現した.しかしながら,周囲の照明条件が変化する環境では,従来手法で目標とする見えに収束させることは難しい.そこで,本研究では動的に変化する環境光を推定し,従来よりも複雑な環境下における見えの制御を提案する.提案手法は,照明条件が変動するモデルを使用し,環境光と白色照明下での見えを過去2回の投影像・撮影像のペアを用いて推定を行う.
 
松宮 翔 1251097: M, 1回目発表 知能コミュニケーション
title: Automatic grant balloon by emotional speech recognition
abstract: A Speech has verbal information and non-verbal information. Non-verbal information includes pitch, energy and emotions and so on. Although current speech recognition exchanges only voice to text, using non-verbal information and verbal information, it performs emotion recognition from speech are proposed. I think of the system that extracts the emotion in real-time using voice recognition, do the clustering in each emotion such as anger or sadness. And I will give video the appropriate balloon. I change the font and the balloon according to the feelings as the manga. I think that convey more meaningful contents to the people who are watching.
language of the presentation:Japanese
 
牛 少博 1251132: M, 1回目発表 ネットワークシステム学
title: Low Complexity Channel Estimation based on Compressed Sensing for OFDM System
abstract: In delay-Doppler double selective channel, the compressed sending (CS) based channel estimation provides accurate channel state information (CSI) and improves BER performance for OFDM systems. However, the high computational complexity due to the big size measurement matrix affects its implementation in high-rate communications. This paper proposes a low complexity scheme of CS-based channel estimation which use a 2-step method to estimate the CSI in time and frequency dimensions separately. The proposed scheme can effectively decrease the computational complexity for reducing the size of measurement matrix. Numerical results demonstrate that the proposed scheme significantly lowers the computational complexity and has good BER performance.
language of the presentation: English
 
久本 空海 1251088: M, 1回目発表 自然言語処理学
title: Parsing the Web with Deep Learning
abstract: Dependency parsing is a core part of natural language processing, and it has seen a great progress in last 10 years. However, the parsing still has challenges in applying it to the non-canonical text such as the ones on the web. In this research we aim to build a robust parser for the web text, using an emerging subfield of machine learning called deep learning, which can handle make use of the unlabeled data. First we will show how the unsupervised word embeddings can help the dependency parsing, and then we will explain how we can extend it to incorporate deep learning into the dependency parsing.
language of the presentation: Japanese
 
田中 勇記 1251059: M, 1回目発表 環境知能学
title: Prediction model for Intelligent Electric Wheelchair
abstract: When pedestrian move, they look at the movement of people around, predict movement of their future, and choose the moving path to avoid collision and dangerous. In related works, The personal space transmutes depending on the target to avoid. In this work, For the robot understands the situation around, modeling to predict the movement of other object. In this presentation, we report on the verification that how to avoid different when the object type is different (e.g. human,wheelchair, and so on)
language of the presentation: Japanese