ゼミナール発表

日時: 6月8日(月)3限 (13:30-15:00)


会場: L1

司会: 候 亜飛
水谷 后宏 1461202: D, 中間発表 藤川和利,安本慶一,猪俣敦夫,砂原(慶應),松浦(東工大),首藤(東工大)
title: A study of the availability improvement of structured overlay network for big data management.
abstract: For the growth of big data analysis' demand, the scalable platform handling the data analysis workload is required in near future. We think structured overlay network technique is one of good solution for handling big data; however, it takes a lot of costs to keep the availability of structured overlay for all that it is good at the scalable data and node management. In this work, we categorize the costs for keeping the availability into three types: data and queries transfer cost, load balancing cost, and maintenance cost. For data and queries transfer cost reduction, we propose multiple data and queries look-up architecture which realizes bulk data and queries transfer. For load balancing cost reduction, we use the overlap ID space management scheme that distributes, fairly, the loads without node migration and with little data reallocation, by sharing some ID-space regions between neighboring nodes. Finally, we apply the testament protocol to reduce maintenance cost. As the comparison evaluation, costs reduction rate of our schemes outperform effective conventional schemes.
language of the presentation: Japanese
 
LI SHA 1451205: M, 1回目発表 ネットワークシステム学
 
里内 匠 1251047: M, 1回目発表 ネットワークシステム学
 

会場: L2

司会: 米田 友和
中川 廣大 1451203: M, 1回目発表 ソフトウェア設計学

title: Agile Development with Software Process Mining[Paper introduction]

abstract:The proper specification and design of software system continues to be a challenge.The majority of current approaches either ignore real user and system runtime behavior or consider it only informally. In this paper, we propose a bottom up approach,which takes event log of software system for the analysis of that.

language of the presentation: Japanese

 
WANG XULIANG 1451211: M, 1回目発表 ソフトウェア設計学

title: *** Flow-level monitoring based-on OpenFlow for categorizing network applications ***

abstract: *** Current network

routing technologies make the route decisions to forward packets of applications

regardless of their different requirements of bandwidth and latency. Which will

make the network work inefficiency. OpenFlow network can take a record of each

flow. We propose a monitoring tool to specify the characteristic of each flow

in flow-level by polling the statistical information of the counter of each

flow, especially the real bandwidth and latency of each flow. This tool will

make a contribution to the controller that is used to optimize network communication.


***

language of the presentation: *** Japanese**

 
PITAKSARIT SIRAWAT 1451209: M, 1回目発表 インタラクティブメディア設計学
title: Testing Environment for Comparing First-Person Video Based and AR Based Learning Method
abstract: Important part of using any system is knowing how to use it. For complex system, augmented reality (AR) aided learning and first-person view instructional video have been proven useful in many situation especially when we can integrate head-mounted display (HMD) into the task then AR or video can be directly displayed overlaid onto user's view. In this study, we wanted to know what kind of situation we prefer AR or video over one another? What are strengths and weakness of both approaches? I am developing a mock-up system that can be used to do this user study.
language of the presentation: English
 

会場: L3

司会: 爲井 智也
近藤 雅也 1451202: M, 1回目発表 ロボティクス
title: Musculoskeletal Hand Model with Individual Differences
abstract: In this study, we developed a subject-specific musculoskeletal hand model considering the differences in the inertia properties and muscle force parameters. Furthermore, we verified the estimated maximum force compared with measured sensor force data while subjects are pinching 28 mm wide object by using developed model. As a result of the simulation, it is suggested that it is possible to perform musculoskeletal simulation considering individual differences.
language of the presentation: Japanese
 
GO CLARK KENDRICK CHENG 1451204: M, 1回目発表 数理情報学

title: Analysis of the Particle Swarm Optimization Algorithm

abstract: Particle Swarm Optimization (PSO) Algorithm was developed by Kennedy and Eberhart in 1995 as a stochastic optimization algorithm inspired by social behaviour and dynamics. In this talk, we review the heuristics for parameter selection to ensure convergence of the algorithm. We also review recent works on the mathematical analysis of the algorithm. Finally, we extend the definition of the state of a particle to include other particles in a swarm.

language of the presentation: English

 
TUCHIN ANDREI EUGENIYEVICH 1451210: M, 1回目発表 大規模システム管理

title: “KnapPath: the centralized management of data centre’s flows”


abstract: In this work, we consider a core algorithm for the centralized management of data center flows. We focus on the arbiter, a centralized controller for scheduling data flows in a data center. The core algorithm of the arbiter controls and distributes flows over the data-center network. In numerical examples, we investigate the flow-allocation rate for several cases of different flow-size distributions. Numerical results show that the proposed algorithm can allocate many flows within a small time interval even when the variance of flow-size is large.


language of the presentation: English

 
PHI VAN THUY 1451208: M, 1回目発表 自然言語処理学
title: Part-Whole Relation Extraction Problem and Word Embedding Approach
abstract: Part-whole relation, or meronymy plays an important role in many domains. Among approaches to address part-whole relation extraction problem, Espresso bootstrapping algorithm proved the effectiveness by significantly improving recall while keeping high precision. We improve Espresso bootstrapping algorithm for part-whole relation extraction problem by integrating word embedding approach into its iterations. The key idea of our approach isutilizing an additional ranker component, namely Similarity Ranker in Instances Extraction phase of Espresso system. This ranker component uses embedding offset information between instance pairs of part-whole relation. The experiments show that our proposed system achieved a precision of 84.9% for harvesting instances of part-whole relation, and outperformed the original Espresso system.
language of the presentation: English
 
PATRICK LUMBANTOBING 1451207: M, 1回目発表 知能コミュニケーション
title: Statistical Approach to Articulatory Controllable Speech Modification
abstract: During our speech production, articulators are moving together to determine the resonance characteristic of the vocal tract. Therefore, speech can be characterized not only by the acoustic parameters, but also by the articulatory parameters. In this work, we develop a speech modification system capable of modifying an input speech signal by manipulating the unobserved articulatory parameters. We do so by sequentially integrating two statistical feature mapping systems based on Gaussian mixture models, acoustic-to-articulatory inversion mapping and articulatory-to-acoustic production mapping. We also implement a speech generation process based on spectrum differential that replaces the vocoder-based process to improve the speech quality. Experimental results show that the system is capable of modifying phonemic sound by manipulating corresponding articulatory positions. Furthermore, higher modified speech quality is yielded from speech generation process based on spectrum differential.
language of the presentation: English