| 室田 勇騎 | 1351108: M, 2回目発表 | 中村 哲, | 杉本 謙二, | 猿渡 洋 | |
|
title: Study on a priori statistical model of target signal for binaural signal source separation
abstract: In music signal separation, it is preferable to retain the sound-localization and reverberation of the separated sound. In order to perform such separation, it is conceivable to use head related transfer functions (HRTFs) such as a binaural information. However the use of this information is not realistic, because this information is normally unknown. To cope with this problem，we focus on the difference between a priori signal distributions at both ears as the binaural information, which can be estimated blindly. Using this information is expected that improve in spatial quality and separation performance of the target sound. In this research, we applied the generalized minimum mean-square error (MMSE) short-time spectral amplitude (STSA) estimator with automatic prior adaptation in a single channel, which have proposed in the past by authors to the binaural signal. Then, improving the separation performance by utilizing the prior distribution of the target sound signal of the left and right channels, which are estimated individually. language of the presentation: Japanese 発表題目: バイノーラル信号音源分離における両耳事前分布モデルの考察 発表概要: 音楽音源分離において，ヘッドホンのように両耳で音を聴取するシステムを考えた場合，音の臨場感などを保つために分離音の定位や残響などは保持されていることが望ましい． このような分離を行うためには，HRTFなどの両耳情報を利用することが考えられるが，HRTFは基本的に未知であるため，実用面においてこれを利用することは現実的ではない。 そこで，ブラインドに推定可能な両耳情報として，目的音の事前分布の違いに着目する。これを両耳の差分として用いることで，分離性能及び再現音の空間的品質の向上につながることが期待される． 本稿では著者らが過去に提案した，シングルチャネルにおける事前分布パラメータ推定を用いた一般化MMSE-STSA推定器をバイノーラル信号に適用する．そして左右チャネルでの目的音信号の事前分布を個別に推定し，それを利用することで，分離性能の向上を図る． | |||||
| 山崎 龍一 | 1251111: M, 2回目発表 | 中村 哲, | 小笠原 司, | 猿渡 洋 | |
| 田中 大介 | 1361008: D, 中間発表 | 杉本 謙二, | 小笠原 司, | 松原 崇充 | |
|
title: Modeling and Control for Active Tactile Object Recognition
abstract: In this presentation, we focus on the active tactile object recognition problem: the informative exploratory action is planned and executed to the object being touched, and the object is recognized by obtained tactile information. To cope with the problem, the following questions should be considered: (1) How to evaluate the informativeness of the action, (2) How to design the compliant exploratory action. For the first question, a previous study indicates that the mutual information (MI) which represents the reduced amount of the uncertainty and is calculated using the observation model can be one of criteria. In order to obtain the model suitable for the object recognition task, we firstly propose to use the unsupervised learning method from the tactile data. Next, since the tactile information is obtained by touching the object, the exploratory action should be compliant as well as informative. In order to design the informative and compliant exploratory action, we secondly propose the optimal control based approach with the constructed observation model. The effectiveness of our proposed methods is validated by showing the experimental results with synthetic data and an actual robot hand. language of the presentation: Japanese | |||||
| 石原 弘二 | 1351006: M, 2回目発表 | 杉本 謙二☆３, | 小笠原 司, | 森本 淳, | 松原 崇充 |
|
title: Model Predictive Control for a Robot with Different Type of Actuators
abstract: In this study, we consider using Model Predictive Control (MPC) to control an exoskeleton robot which has different type of actuators. Since joint angle movements of the exoskeleton robot can be disturbed by physical interactions with users and mathematical models of the robot and the actuators have modeling errors, we need an optimal feedback policy for exoskeleton robot control to cope with the disturbances. However, it is generally difficult to analytically derive an optimal controller for a high-dimensional nonlinear system such as an exoskeleton robot. MPC can deal with the disturbances because it generates a locally optimal control input as an optimal feedback policy through online trajectory optimization. Due to the recent rapid improvements in computer performance, MPC is becoming a popular approach to generate optimal movement trajectories in robot control. However, MPC has been considered as a computationally intensive method and difficult to be applied to a robot with different type of actuators. In this talk, we propose a computation method reducing the computation time of optimization in MPC to control the robot which has different type of actuators and investigate how the proposed MPC works to generate an optimal feedback policy on a single joint arm robot in trajectory tracking tasks. We show that the proposed method reduces the computation time and achieve better tracking performances than conventional MPC. language of the presentation: Japanese | |||||
| 中山 直輝 | 1351080: M, 2回目発表 | 飯田 元, | 松本 健一, | 市川 昊平 | |
|
title: Investigation of defect rate within code clone and the surrounding code.
abstract: Code clone is a duplicate code fragment in the source code of software and generated by developer’s copy and paste. Recently, some researchers proposed defect detection method based on code clone analysis and actually detected some software defects from some projects. However, detecting defect based on code clone analysis may require developer to spend effort to check whether there is any defect because defect may exist within the code clone or codes surrounding it. By investigating the defect rate in the code clone and their surroundings, developers can efficiently prioritize the area to check. Furthermore, if the investigation reveals that the surrounding code has higher defect rate than within the code clone, another investigation will be conducted to find out which particular surrounding code has high defect rate. By doing so, I hope to increase the efficiency of checking. language of the presentation: Japanese 発表題目: コードクローン内外における欠陥含有率の調査 発表概要: コードクローンとは，ソースコード中の互いに一致または類似したコード片を指し，主に開発者が行うコピーアンドペーストによって発生する．近年，コードクローン解析に基づくソフトウェアの欠陥検出手法が提案されており，実際に欠陥を検出した事例も報告されている．しかし，コードクローンに着目して欠陥を探す際に，欠陥がコードクローン内に含まれる場合もあれば，コードクローン外に存在する場合もあるため，開発者はコードクローン内外の両方について欠陥の有無を調べなければならない．そこで本研究では，オープンソースソフトウェアを対象にコードクローン内外における欠陥含有率を調査し，開発者がコードクローン内外のどちらを優先的に確認すべきかを明らかにする．また，欠陥がコードクローン外に多く存在する場合，欠陥含有率が高いソースコードの範囲を調査することで，開発者が欠陥の有無を確認する際の作業効率化を実現する． | |||||
| KANASHIRO PEREIRA LIS WEIJI | 1361016: D, 中間発表 | 松本 裕治, | 松本 健一, | 新保 仁, | Kevin Duh |
|
title: Automated Lexical Choice Error Correction for Second Language Learners abstract: Error correction in non-native writing has been a focus of research in recent years. However, research this area usually focuses on correction of function words (e.g. prepositions, articles, pronouns), while correction of content word errors (i.e. noun, verb, adjective and adverb) or lexical choice errors have not been explored extensively. We report our experiments to correct one type of lexical choice errors, i.e. collocation errors, which are commonly found in non-native writing. While previous works in the literature used first language (L1) text for generating correction candidates, we investigated using a large Japanese language learner corpus to build a system that is more sensitive to constructions that are difficult for learners. Our results showed that by using this learner corpus, we obtained better results compared to other methods that use only L1 text. Furthermore, we propose an extension of the current method to handle the other types of lexical choice errors. language of the presentation: English | |||||
| 小林 靖幸 | 1351042: M, 2回目発表 | 伊藤 実, | 松本 健一, | 楫 勇一, | 関 浩之 |
|
title: Quantitative Evaluation of the Key Information That is Leaked through Timing Attack for RSA Cryptosystem
abstract: Timing attacks are regarded as serious threats to many practical cryptographic algorithms, but it is difficult to estimate the risk of the attack. In previous research focuses on the mutual information between a secret key and timing observations that an attacker acquires. The discussion of this approach gives a certain upper-bound on the risk of timing attack. In my presentation, I would like to explain the upper-bound of the mutual information on the entropy of multinomial distribution that is tighter than the conventional method. language of the presentation: Japanese 発表題目: RSA暗号に対するタイミング攻撃により漏えいする鍵情報の量的評価 発表概要: タイミング攻撃は，暗号アルゴリズムの実装に対して深刻な脅威となりうる．しかし，タイミング攻撃のリスクを定量的に評価することは困難である．先行研究として，RSA暗号に対して攻撃者が入手する実行時間と鍵情報との間の相互情報量をリスクの定量的指標として採用している．本発表では，多項分布を用いたエントロピーに関する結果を導入することで，先行研究の改善について述べる． | |||||