ゼミナール発表

ゼミナール発表

日時：	平成22年5月31日(月)3限 (13:30 -- 15:00)

場所：	L1
司会：	橋本

AHMED ASAAD AHMED GAD EL-RAB 　D	ソフトウェア基礎	伊藤実	関浩之	安本慶一
題目：Efficient Methods for Increasing Service Availability and Spatio-temporal Sensing Coverage in Mobile Ad-hoc Networks
概要：In this presentation, we will propose two methods for realizing dynamic service replication in mobile ad hoc networks (MANETs) and on-demand urban sensing utilizing pedestrian MANETs. Improving service availability in MANETs is a challenging task. However, service replication can considerably impact the system energy consumption. Since mobile devices have a limited amount of battery, a dynamic and efficient service replication is necessary. In the first part of this presentation, we propose a distributed service replication scheme for achieving high service availability with reasonable energy consumption for MANETs. The proposed method called HDAR (Highly Distributed Adaptive Service Replication) divides the whole network into disjoint zones of at most 2-hops in diameter and builds a dynamic replication mechanism which selects new replica zones depending on their service demand and the tradeoff between the communication and replication energy consumption costs. Through simulations, we have confirmed that our approach can achieve higher service availability with reasonable energy consumption independently of network size than existing methods. There is an increasing demand for obtaining environmental information of the specified region in the urban district for various purposes such as surveillance, navigation, and so on. Pedestrians with mobile phones can be used as mobile sensor nodes to cover given areas of interest (AoI) over time. In the second part of this presentation, we propose a concept of spatiotemporal coverage by a set of mobile sensor nodes for a given AoI, possibly under deadline constraints. We model an urban sensing scenario with pedestrians as mobile sensor nodes moving according to a discrete Markov model and present probabilistic algorithms based on this model to improve the total coverage under deadline constraints.

ZAINAL　ARIEF 　D 生命機能計測学湊小太郎　木戸出正繼杉浦忠男佐藤哲大
題目：Fuzzy rule-based classifier in medicine application.

概要： The important task in fuzzy classification system is to define the proper set of fuzzy rules which can be built by data driven or expert knowledge to assigns a class label to an object. In this work, we try to use the mean and standard deviation of measured variables approach to built fuzzy if-then rule. The mean and standard deviation of measured variables are derived from existing data to assign class label based on the object description. The first application of the scheme is classification between fatigue and non fatigue condition using extracted parameters from eye movement data as a Gaussian shape fuzzy sets. Each condition is assigned from one rule in antecedent with saccadic latency, duration, velocity and deviation as the induced fuzzy sets. The classification accuracy of 86.54% is achieved as a result. The second application is cardiac rest period determination from magnetic resonance of cardiac image. The cross-correlation value of consecutive image and normalized frame number are treated as measured variables to define the fuzzy sets. Two radiologist decisions are modeled in this work. The results show that the fuzzy classifier system can effectively characterize the radiologist decision. From the two application results, the merit of fuzzy classifier is shown in term of its simplicity and accuracy.

間所洋和　D 知能情報処理学木戸出　正繼小笠原　司浮田　宗伯
題目：視覚移動ロボットのための教師なしカテゴリ分類 (Unsupervised Category Classification for a Vision-Based Mobile Robot)

概要：オフィスや家庭などの一般的な環境において，ロボットが自律的かつ合目的に行動するためには，周囲の状況を能動的に認識しながら概念世界のパターンとなる世界像を形成することが求められる．本研究では，ロボットの移動に伴う見え方の変化から，視野画像列の分類により自己位置を推定し，環境内の一般物体を認識する手法を提案する．提案手法では，安定性と可塑性を併せ持つ適応共鳴理論ネットワークを用いてカテゴリの候補となるラベルを生成し，競合と近傍に基づく写像特性を有する自己組織化マップを用いてカテゴリとして分類する．実ロボットを用いた評価実験から，実環境におけるカテゴリ分類の有効性を示すとともに，世界像形成のための記憶パターンをカテゴリマップとして表現できることを示す．

ゼミナール I, II ページへ

ZAINAL　ARIEF 　D	生命機能計測学	湊小太郎	木戸出正繼	杉浦忠男	佐藤哲大
題目：Fuzzy rule-based classifier in medicine application.
概要： The important task in fuzzy classification system is to define the proper set of fuzzy rules which can be built by data driven or expert knowledge to assigns a class label to an object. In this work, we try to use the mean and standard deviation of measured variables approach to built fuzzy if-then rule. The mean and standard deviation of measured variables are derived from existing data to assign class label based on the object description. The first application of the scheme is classification between fatigue and non fatigue condition using extracted parameters from eye movement data as a Gaussian shape fuzzy sets. Each condition is assigned from one rule in antecedent with saccadic latency, duration, velocity and deviation as the induced fuzzy sets. The classification accuracy of 86.54% is achieved as a result. The second application is cardiac rest period determination from magnetic resonance of cardiac image. The cross-correlation value of consecutive image and normalized frame number are treated as measured variables to define the fuzzy sets. Two radiologist decisions are modeled in this work. The results show that the fuzzy classifier system can effectively characterize the radiologist decision. From the two application results, the merit of fuzzy classifier is shown in term of its simplicity and accuracy.

間所洋和　D	知能情報処理学	木戸出　正繼	小笠原　司	浮田　宗伯
題目：視覚移動ロボットのための教師なしカテゴリ分類 (Unsupervised Category Classification for a Vision-Based Mobile Robot)
概要：オフィスや家庭などの一般的な環境において，ロボットが自律的かつ合目的に行動するためには，周囲の状況を能動的に認識しながら概念世界のパターンとなる世界像を形成することが求められる．本研究では，ロボットの移動に伴う見え方の変化から，視野画像列の分類により自己位置を推定し，環境内の一般物体を認識する手法を提案する．提案手法では，安定性と可塑性を併せ持つ適応共鳴理論ネットワークを用いてカテゴリの候補となるラベルを生成し，競合と近傍に基づく写像特性を有する自己組織化マップを用いてカテゴリとして分類する．実ロボットを用いた評価実験から，実環境におけるカテゴリ分類の有効性を示すとともに，世界像形成のための記憶パターンをカテゴリマップとして表現できることを示す．