ゼミナール発表

日時: 12月14日(水)3限(13:30-15:00)


会場: L1

司会: 小林 泰介
LIU CHENGBO 1551203: M, 2回目発表 ネットワークシステム学 岡田 実,杉本 謙二,東野 武史,侯 亜飛

title: Subcarrier index modulation based flexible OFDM system

abstract: OFDM is an attractive system due to its high spectrum efficiency and high tolerance to multipath interference. However, the system has large peak-to-average ratio (PAPR) which requires expensive and large size power amplifier. On the other hand, some small devices usually can be designed with a simple structure if they select low-order M-QAM modulation like QPSK to transmit data. Therefore, the OFDM system will be more promising if it can descend the order of M-QAM modulation with almost identical bandwidth efficiency (BWE) compared to that of original OFDM system and reduce the PAPR. In this research, we will combine OFDM system with subcarrier index modulation (SIM) and propose two systems which are called half-symbol BPSK modulated SIM-OFDM (HS/SIM-OFDM) and double ASK-SIM for OFDM system, respectively. The former system can double the BWE but require high complexity due to M-algorithm. The latter system can double the BWE with a considerable complexity. Moreover, both proposed systems can achieve better PAPR performance than that of original OFDM system with comparable BER performance. The proposed systems can achieve flexible BWE with low-order M-QAM modulation and better PAPR characteristic with comparable BER performance.


language of the presentation: English

 
鶴峯 義久 1651073: M, 1回目発表 知能システム制御 杉本 謙二
title: Deep Reinforcement Learning based on Dynamic Policy Programming
Deep Reinforcement Learning (RL) has been drawing much attention since it enables to learn control policies of agents from raw images. However, it requires a huge number of training data. To alleviate this issue, previous studies have focused on improving the network architecture. In this research, we explored another direction: propose a novel deep reinforcement learning that combines dynamic policy programing and deep neural network. In this presentation, we demonstrate the effectiveness of our method as compared to a previous method for a robot arm control problem in simulation.
language of the presentation: Japanese
発表題目: 動的方策計画に基づく深層強化学習
発表概要: 深層強化学習による画像からの行動学習が注目されている。 しかし、この様な行動学習は学習に大量のデータが必要なため、 先行研究ではネットワークアーキテクチャを工夫することで必要データ数の削減が試みられている。 本研究では方策を少しずつ更新する動的方策計画と深層学習を組み合わせ、 比較的少ないデータ数で学習できる手法を提案する。 提案手法と従来手法を学習に必要なデータ数で比較し、提案手法の有用性を示す。
 
真鍋 陽俊 1651100: M, 1回目発表 自然言語処理学 松本 裕治
title: Improvement of pipelined system in natural language processing using FFNN-Filtering model
abstract: Many natural language processing tasks have multiple components. In dependency parsing, for example, we usually use predicted part-of-speech(POS) tags as features. However, such pipelined systems suffer from error-propagation; the error caused in low-level tasks cascades into the error in higher-level tasks. In previous researches, joint modeling methods are often proposed to deal with this problem, but these methods increase search space. In this research, we propose Filtering model which can prevent error-propagation problem and reduce search space.
language of the presentation: Japanese
 

会場: L2

司会: 樫原 茂
柴田 大作 1651059: M, 1回目発表 知能コミュニケーション 中村 哲☆
title: Study on detecting Alzheimer's Disease using Natural language process
abstract: In recent years, detecting Alzheimer’s disease (AD) in early stages based on natural language processing (NLP) has drawn much attention. To date, vocabulary size, grammatical complexity, and fluency have been studied using NLP metrics. However, the content analysis of AD narratives is still unreachable for NLP. This study investigates features of the words that AD patients use in their spoken language. In this presentation we propose method to classify words into categories detecting AD and evaluate it.
language of the presentation: Japanese
発表題目: 自然言語処理による認知症スクリーニングの検討
発表概要: 自然言語処理による認知症の早期スクリーニングが注目されている。現在まで、語彙サイズ、文法の複雑さや言語流暢性などに注目した研究が報告されている。しかしこれらは、患者の発話を1つの統計量として扱っており、問題を過剰に簡単化している恐れがある。そこで我々は患者の発話内容に踏み込んだ質的な観点による認知症スクリーニングの手法の検討を行う。本発表では語彙をカテゴリ分類することによる認知症スクリーニングの手法を提案し、その評価を行う。
 
小牧 真子 1651048: M, 1回目発表 知能コミュニケーション 中村 哲
title:Fast computation of planning optimal farm works
abstract:The aim of this study is for reducing the computational cost to fix the plan of farmwork . When the worker devises a farming plan, we should consider the number of fields, kinds of works and working time.Therefore, when we calculate the most optimal farming plan, we should consider all combinations of farming works, then we should take huge amount of calculation time.To reduce the calculation cost, we introduce a technique for generating constraints automatically from the temperatures of each farm and the growth degrees of each crop, which is used for dynamic programming.
language of the presentation: Japanese
発表題目:農作業計画の高速計算にむけて
発表概要:本研究では,農作業計画を策定するために必要な計算量の削減を目的とする. 農作業計画を策定する際,圃場数,作業の種類数や作業時間など様々な要素を考慮する必要がある. そのため,最適な計画を求めるためにはこれらの要素の組合せ数が膨大となり計算時間がかかることは問題となる. そこで動的計画法を用いる際に農作業計画で考えなければならない作物の成長度合いや天候などの要因から考慮不要な要素の組合せを取り除くことにより計算量を削減することができると考えた.
 
加藤 涼子 1651033: M, 1回目発表 インタラクティブメディア設計学 加藤 博一
Title:(Paper Introduction) Interaction-Free Calibration for Optical See-Through Head-Mounted Displays Based on 3D Eye Localization(INDICA)
Recently, Head-Mounted Displays(HMD) are widely used in our daily life. In particularly, Optical See-Through(OST) HMD is used for Augmented Reality(AR) Aplications, and this aplication require properly calibration so as to 2D point on the virtual scene corresponds to 3D point on the real scene correctly, and SPAAM calibration method requires repeatedly user interaction in calibration procedure. However, repeatedly calibration task impose a great burden on user, and in this calibration procedure, OST-HMD can not move on user's head. In this paper, authors proposed automated calibration method for OST-HMD using eye localization, and INDICA not require user interaction. In today's presentation, I will introduce two OST-HMD calibration method; SPAAM and INDICA.
language of the presentation:Japanese
 
杉山 玲央奈 1651062: M, 1回目発表 ネットワークシステム学 岡田 実
title:kQ-product Analysis of Multiple-Receiver Inductive Power Transfer with Cross-Coupling
abstract:This resarch investigates maximum achievable efficiency of inductive power transfer (IPT) system with arbitrary number of receivers. We derive the formulas of optimal loads analysis and computer simulations, we show that the cross-coupling among receivers does not affect system efficiency if load resistances and reactnaces are jointly optimized. We also prove that the maximum efficiency is donated by a so-called system kQ-product, whose square is interestingly, equal to the sum of squares of kQ-product of individual transmitter-receiver links.
language of the presentation:Japanese
 
盒 雄太 1651067: M, 1回目発表 ユビキタスコンピューティングシステム 安本 慶一

発表題目:労働生産性改善に向けたウェアラブル機器を用いた体調推定

発表概要:体調か優れない状態て労働した場合の労働生産性の低下か問題視されており,経済的損失か大きいことかわかっている.

企業の経営者側としては,従業員の健康状態を把握・改善することて,潜在的な経済的損失を減らすことかてきる.

また,労働者側としては,自己の体調を把握し,生活を改善することかてきれは,体調か良い状態て仕事に取り組むことかてきる.

そこで,労働時以外の生活リスムや生体・行動テータを取得てきることから,ウェラフル機器に着目し,ウェアラフル機器を用いた労働生産性を改善するシステムの構築を目指す

本発表では,労働生産性に影響のある体調を選定し,その体調がどのようなセンサで計測できるか,調査結果を報告する.

Title:Estimation of physical and mental condition by wearable devices for improving productivity.

abstract:It is known that working on a bad physical/mental condition reduces productivity and causes a lot of economic loss.

As a manager of a company, it is possible to reduce a potential economic loss by grasping by improving a health condition of employees.

Also, as a worker of a company, if he/she can grasp his/her physical/mental condition, it helps improving his/her life.

In this study, we focus on wearable devices to sense life rhythms, biological data and behaviors, aiming to develop a system that estimates the physical/mental condition and improves productivity using those devices.

In this presentation, we selected physical and mental conditions that affect productivity and report a result of surveying what sensors can measure each of those conditions.

language of the presentation:Japanese

 
豊島 健太 1651079: M, 1回目発表 ロボティクス 小笠原 司

title: Cable and connector manipulation for parts replacement

abstract: Many researches on automation of production activities by robot arms are carried out in research in the industrial technology field. However, automation of disassembly work has not been done much. Disassembly work is a technique necessary for repairing and checking products and for recycling to take out resources from end of life products. but now most of them are done by humans. In this presentation, we will introduce the difficulties of automation of disassembly work, research to assist disassembly of batteries of electric vehicles by robot, research on manipulation targeting cables which becomes a problem in disassembling products, and describe future research.

language of the presentation: Japanese

発表題目: 部品交換のためのコネクタケーブルマニピュレーション

発表概要: 産業技術分野の研究において,ロボットアームによる生産活動の自動化に関する研究は多く行われている.しかし,分解作業の自動化はあまり行われていない.分解作業は製品の修理や点検を行う際や,寿命を迎えた製品から資源を取り出すリサイクルを行う際に必要な技術であるが,現在はほとんど人の手によって行われている.本発表では分解作業の自動化の難しさ,ロボットによる電気自動車のバッテリの分解の補助を行う研究,製品を分解する際に問題となるケーブルを対象としたマニピュレーションについての研究の紹介を行い,今後の研究について述べる.

 

会場: L3

司会: Doudou Fall
福岡 久和 1651091: 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
title: EMAXV: Low Power and High Performance CGRA Accelerator
abstract: According to the senior white paper, in 2025, one third are seniors. Care-support systems with machine learning, such as Deep Convolution Neural Network (DCNN), has been developed. Some kind of computing devices, such as CPU, FPGA, and GPU, cannot achieve high performance while keeping low power consumption. Nowadays, CGRA accelerators are paid more attentions in order to balance performance and energy efficiency. Therefore, currently we are developing CGRA accelerators, called EMAXV, on FPGAs. However, our final target is to implement EMAXV on ASIC. In this report, we present our research progress of prototyping EMAXV on FPGAs.
language of the presentation: Japanese
 
西川 雅清 1651082: M, 1回目発表 光メディアインタフェース 向川 康博
title: Visualizing Ray Refraction Quantity depending on Angle of Cell Surface
abstract: (特許申請のため非公開)
language of the presentation: Japanese
発表題目: 細胞球面との入射角度に応じて生じる光線の屈折量の可視化
発表概要: (特許申請のため非公開)
 
原 崇徳 1651089: M, 1回目発表 大規模システム管理 笠原 正治
title: Path selection based on road reliability for evacuation guiding
abstract: When large-scale disasters occur, evacuees have to evacuate to safe place quickly and safely. In this presentation, we propose a short and reliable path selection for evacuation guiding, which prevents evacuees from encountering blocked road segments as much as possible. First, the proposed scheme calculates k-shortest paths from the current location to the destination, with the help of the existing algorithm. Then, it selects the most reliable one from the candidates by taking account of road blockage probabilities, each of which is an estimated probability that the corresponding road is blocked under a certain disaster. In recent years, we can obtain such information about road blockage probabilities from some autonomy, e.g., Nagoya city in Japan, in advance. Through simulation experiments, we show that the proposed scheme can reduce the number of encounters with blocked road segments with appropriate value of k, while keeping the average evacuation time compared with the shortest path selection.
language of the presentation: Japanese

 
野添 光 1651086: M, 1回目発表 数理情報学 池田 和司
Title:A stochastic model of clock gene related to somite segmentation
abstract:One of the most important life phenomenone, somite segmentation is confirmed in a growth process from zygote to organism. Somite segmentation divide cell population into somite which forms space period. Previous research discovered a relationship between space period to time period of clock gene, and time period is converted to space period. But, somite segmentation has robust system toward space perturbation, but previous mathematical model don‘t reprocut this robust system. This research focuses on creating mathematical model which has robust system based on biological knowlege.
発表題目:体節形成に関わる時計遺伝子の確率モデル
発表概要:受精卵から生体に成長するまでの過程で, 生物に共通して確認される重要な生命現象の一つに体節形成が存在する. 体節形成は, 細胞集団を体節に分割し, 空間的な周期性を作る. 先行研究ではこの空間的な周期性は時計遺伝子の周期的な発現と関係していることが知られており, 時計遺伝子の時間情報が体節の空間情報に反映される. しかし, 実際の生物の体節の空間的な周期性は頑強な同期同調性を備えているが, その頑強性は既存の数理モデルでは再現できていない. 本研究では既存の数理モデルより生物的知見に基づいた形で数理モデルを構築し, 体節形成の頑強性を再現することを目的とする.
 
中谷 聡志 1651081: M, 1回目発表 生体医用画像 佐藤 嘉伸

title: Development of automated organ segmentation method from the endoscopic image in laparoscopic surgery

abstract: In laparoscopic surgery for lymph node dissection in the treatment of gastric cancer, it is important to estimate the position of the pancreas and to determine the region of fat that requires dissection. However, pancreas is covered with fat. Therefore, estimating its position and determining the region of fat is difficult for inexperienced surgeons. In this research, we aim at automatic segmentation of organs in endoscopic images. In this presentation, I present the result of segmentation using the random forest for the endoscopic image of an actual case.

language of the presentation: Japanese

発表題目: 腹腔鏡下手術における内視鏡画像からの臓器自動セグメンテーション手法の開発

発表概要: 胃がん治療などにおけるリンパ節郭清のための腹腔鏡下手術では,内視鏡画像中における膵臓位置の推定や郭清対象となる脂肪領域の判断が重要となる.しかしながら,術中の膵臓は脂肪に覆われているため,膵臓位置の特定や,郭清対象とする脂肪の判断は熟練医でなければ難しい.そこで,本研究では内視鏡画像中の臓器に対しての自動セグメンテーションを目指す.今回の発表では,実症例の内視鏡画像に対してランダムフォレストを用いたセグメンテーション結果について報告する.