ゼミナール発表

日時: 12月11日(火)3限 (13:30-15:00)


会場: L1

司会: 油谷 暁
伊原 啓晃 1251013: M, 1回目発表 ユビキタスコンピューティングシステム
title: Dinner recipe recommendation based on among preferences, nutrition, and ease of cooking
abstract: In this study, we propose a method to recommend a dinner recipe. The decision of the dinner recipe is troublesome since we need to consider (1) too many candidates of recipes, (2) many factors, (3) many times of selections. Therefore, it would be useful if we could have a system that supports the decision making of the recipe. In this study, assuming that the main factors to decide dinner recipes are preferences, nutrition, ease of cooking, budget, and expiration date of foodstuff, we aim to realize the system to recommend the most suitable dinner recipe taking into account the balance of these factors. There are two main problems to construct this system: One is considering five factors to recommend dinner recipes, the other is finding the best weight on each factor to cope with the sense of values different from person to person. To solve these problems, we propose a method for scoring each factor for various dinner recipes, and optimizing weights on the factors to match the sense of values of the user. In this presentation, we describe the investigation of the above issues and the current progress.
language of the presentation: Japanese
発表題目: 嗜好・栄養・調理の簡単さ等をバランスよく考慮する夕食レシピ推薦法
発表概要: 本研究では,夕食レシピ推薦システムの提案を行う. 夕食レシピの決定は、(1)選択肢が多い、(2)考慮すべき要因が多い、(3)選択回数が多い、という点で面倒であり,レシピの決定を支援してくれるシステムがあれば便利である.本研究では,嗜好・栄養・調理の簡単さ・消費期限・食費の5つがおもなレシピ決定の要因であると想定し,これら要因をバランスよく考慮して最適なレシピを推薦するシステムの実現を目指す.このシステムの実現にあたり,上記の5つの要因を考慮することと,ユーザごとの異なる価値観に対応するために各要因にかける重みを動的に変更できるようにすることが課題となる.本課題を解決するため,各レシピに対する各要因の数値化の方法と,各要因の数値にかける重みをユーザの価値観に合わせて最適化する方法を提案する.本発表では,上記の課題の検討とその進捗について述べる.
 
中川 尊雄 1251075: M, 1回目発表 ソフトウェア工学
title: Program Comprehension Process Measurement Based on Cerebral Blood Flow
abstract: Program comprehension is required in each step of the software development such as code review, coding, software maintenance and operation. Many studies that analyze the program comprehension process have been performed to clarify how developers comprehend the program or where developers confused in. A lot of measurement methods have been proposed. For example, recording operation history of the development environment and obtaining the eye-gaze tracking data. My research goal is to measure the brain activation during the program comprehension process by the subject experiment using the cerebral blood flow measurement device.
language of the presentation: Japanese
発表題目: 脳血流量に基づくプログラム理解行動の計測
発表概要: プログラムの理解は、コードレビューやコーディング、ソフトウェアの保守運用などソフトウェア開発の各工程で必要とされる。開発者がプログラム理解を行う過程や、プログラムのどのような部分で理解しがたく感じているかなどを知るため、プログラム理解行動の計測が行われている。プログラム理解行動の計測手法には、開発環境の操作履歴や、コードレビュー中の視線計測データを取得するものなどが提案されている。本研究では、プログラム理解行動中の開発者の脳活動を計測するため、脳血流量測定装置(NIRS)を用いた被験者実験を行う。
 
畑中 健作 1251083: M, 1回目発表 情報基盤システム学
title:MANET/DTN Switching for Wireless Multihop Networking in Realistic Environment
abstract:The infrastructureless communication system is required for severe environment.The Wireless Multihop Network is major research area for the infrastructureless communication system. But the state-of-the-art Protocols are designed for narrow target environment.Therefor We propose MANET/DTN Switching which perform high packet delivery ratio in realistic environment.
language of the presentation: Japanese
 
中島 悟 1251077: M, 1回目発表 ディペンダブルシステム学
title:Paper introduction:Chasing the Weakest Failure Detector for k-Set Agreement in Message-passing Systems
abstract: It is know that many problems cannot be solved in asynchronous message-passing systems prone to any number of process failures. The k-set agreement problem which is one of the fundamental problems in distributed systems is a unsolvable problem. This problem generalizes a consensus problem. The condition of the failure detector which allows the k-set agreement problem to be solved in asynchronous message-passing systems has been studied. Failure detectors are distributed oracle that provide each process with information about failures of other processes. In particular, the weakest failure detector provide each process with necessary and sufficient information required in order to solve problems. In this paper, the new failure detector which is weaker than known failure detectors so far is proposed. and it is proved that solves the k-set agreement problem in asynchronous message-passing systems.
language of the presentation: Japanese
 
陳 少鵬 1251063: M, 1回目発表 計算メカニズム学

title: User Authentication Utilizing Amida-Kuji

abstract: Practical user authentication methods often make use of by primitive belongings such as password-cards . However, the password-cards are not secure enough and many problems have been reported so far. To replace for password-cards, we consider a user authentication method utilizing "Amida-Kuji" which is a traditional lottery of Japan. In the talk, we introduce our preliminary investigation on the statistical properties of Amida-kuji, which will be the security basement of the proposed authentication method.

language of the presentation: Japanese

発表題目:「阿弥陀籤」を使用した個人認証方式の提案

発表概要: 計算機能を持たない所有物を併用した個人認証方式として,暗証カードを利用した方式がオンラインバンク等で広範に利用されている.しかし,その仕組み上,暗証カードの安全性はきわめて低く,実際に不正アクセス等の問題が多数発生している.そこで本研究では,暗証カードの代替とすべく「阿弥陀籤」を利用した認証方式の検討を行なっている.本発表では,阿弥陀籤の持つ統計的性質や暗号学的強度に関する基礎的な検討結果を紹介する.

 

会場: L2

司会: 河合紀彦
常松 佑太 1251066: M, 1回目発表 視覚情報メディア
title: Paper introduction: Single-View Sketch Based Modeling
abstract: abstract: Today, many sketch-based modeling systems have been proposed, but most of them require some operations too different from actual drawing methods. In this paper, a new sketch-based modeling system which uses an actual drawing method as a reference is proposed. This system allows the user to make 3D models from a single viewpoint in the way close to drawing. Because this system aims at allowing user to come up with new ideas for design, the system generates rough 3D models. However, the kinds of shapes that this system is able to generate are enough for the aim of this system.
language of the presentation: Japanese
 
MD.SALIK PARWEZ 1251128: M, 1回目発表 ネットワークシステム学

title: Channel Estimation in Radio over Fibre Distributed Antenna System

abstract: To meet the increasing demand of high speed internet, high definition videos, exchange of large volume data and other high throughput requiring services, Radio over Fiber Distributed Antenna System(RoF-DAS) has been a viable choice nowadays. In RoF-DAS, the multiple antennas (Radio Remote Unit, RRU) are distributed at different location but they communicate with the user in cooperative way which is regulated at the back end by a Central Control Station (CCS). To do so, CCS needs to know the channel behavior (Channel State Information, CSI)between the transmitting stations and the mobile user. Since the channel between the transmitter and receiver are contained with different kind of terrains, buildings, moving vehicles, etc., causes the channel deteriorate and hence very difficult to estimate, which further causes problem in providing the desired capacity and coverage. Moreover, as the number of RRUs and the number of antennas on the mobile terminal increases, the channel becomes more complex causing delay in signal processing and performance degradation. In this scenario of large number of RAUs and multi-antenna users, I am working to reduce the complexity of these channel as well as I intend to propose pilot distribution method to estimate the channel more efficiently in case of RoF-DAS.

language of the presentation: English

 
于 水 1251131: M, 1回目発表 知能コミュニケーション
 
小林 和也 1251045: M, 1回目発表 自然言語処理学
title: Post-ordering in Japanese-English Machine Translation
abstract: Statistical Machine Translation does not work well when word orders are completely different such as Japanese and English. (Sudo et al., 2011) proposed the post-ordering framework for Japanese to English translation. The post-ordering framework translates Japanese sentences into Japanese-ordered English at first, and then reordered into correctly-ordered English. In this presentation, I propose the way to improve the reordring method and show results of fundamental experiments.
language of the presentation: Japanese
 
松山 洋一 1251099: M, 1回目発表 環境知能学
title: Pose estimation in still images
abstract: There are large numbers of various images and videos in the Internet, TV shows, personal albums, and surveillance cameras. For quick and correct retrieval, those videos and images should be tagged with the human actions. However, since the number of videos and images has increased, tagging all of them is difficult. To solve this problem, an automatic tagging teshnique using human action recognition in videos and images. I present this issue and how to achive pose estimation in still images for the basis of the human action recognition.
language of the presentation: Japanese
発表題目:静止画における人物の姿勢推定
発表概要: 近年,インターネット上の動画や画像,テレビや個人のアルバム,監視カメラなど様々な動画や画像が数多く存在している.これらの動画や画像を高速に検索するためには,その中に映っている人物の行動によってタグ付けをする必要がある.しかし,その数が増え続けているため手動によるタグ付けは困難となってきている.そのため 動画や画像内での人物の行動を認識し自動的にタグ付けを行う技術が必要とされている.本発表では人物の行動認識の基礎技術となる静止画上での人物の姿勢を推定する方法や課題について述べる.
 

会場: L3

司会: 竹村憲太郎
椿 真史 1251067: M, 1回目発表 数理情報学
title: Distributed Representation with Neural Language Model and Semantic Space Model for polysemous words
abstract: Semantic single-word vector space models have been very successful at many natural language processing (NLP) tasks and applications such as information retrieval and document classification. Despite their success, one major limitation is that these representations have only one representation for each word. This representation method cannot capture the different meanings of word depend on contexts (multi-sense words). In this presentation, I propose the model to capture the change of word meanings in the vector space model with Distributed Representation by Neural Language Model.
language of the presentation: Japanese
 
FOIMAE Ori Amao 1251125: M, 1回目発表 知能システム制御
title: Comparision of I-PD control and PID control with model based robust tuning for PID parameters.
abstract:In production of many consumer goods, the need for highly stable operations is a key objective to achieving production targets. In process control, several methods are utilized to achieving such targets; one example is in the better utilization of PID control, where a variation of the PID controller, namely the I-PD controller is used to smoothen out the effect of abrupt responses to set point changes. In this seminar presentation I will compare the PID controller with the I-PD controller presentated by Kano et al in メThe state of the Art in Advanced Chemical Process Control in Japanモ and present a Model based Robust PID tuning for the I-PD controller by Ogawa et al.
language of the presentation:English
 
馬場 浩平 1251084: M, 1回目発表 ロボティクス
title: Generating a variety of robot behaviors by capturing human motion
abstract: Recently, many mobile service robots have been developed, and we expected that the robots work in the real world with human. However, most of the robots provide behaviors, which were programed in advance. It is necessary that robots acquire more flexible and various behaviors like a human. In this presentation, I introduce the related works for motion primitive, and I propose that robots get various behaviors from human by using a sensor suits.
language of the presentation: Japanese
 
林 平将 1151086: M, 1回目発表 ロボティクス
title: Rough terrain clustering using walking policy in multi-legged robot
abstract: In this research, we will propose rough terrain clustering algorithm using walking policy robots already own. First of all, we make some types of rough terrains and generate walking policies for each terrain by reinforcement learning. Based on rewards getting from adapting each walking policy, we cluster these terrains into some groups. After that, we decide the most robust walking pattern in each group. Adapting these gaits into unknown rough terrains not used at training time, we estimate their groups of rough terrains from their rewards.
language of the presentation: Japanese
 
内藤 諒 1251071: M, 1回目発表 知能システム制御
title: Modeling of antagonistic Mckibben pneumatic artificial muscle system and experimental validation
abstract: A McKibben pneumatic artificial muscle is expected to be used for assistive products. But, the realization of position control and force control about a McKibben pneumatic artificial muscle system is difficult problem due to very complicated non-linear characteristics. It's needed that mathematical model to grasp output of pressure and contraction ratio for input voltage when controlling position and force. In our previous work, we proposed a nonlinear modeling method McKibben artificial muscle system under load changes. In this presentation, I'll show the idea that applies this model to modeling a antagonistic McKibben pneumatic artificial muscle system and introduce the previous work.
language of the presentation: Japanese