The 2nd COE Postdoctoral and Doctoral Researchers
Technical Presentation

Date: Thursday, May 25, 2006
Time: 13:30 - 16:20
Place: L1 Lecture Room
Language: English ( Oral Presentation ), English/Japanese ( Discussion )
Chairperson: Muneyuki Sakata ( Image Processing Lab. : COE assistant professor ),
Carlos Troncoso Alarcon ( Speech and Acoustics Lab. : COE assistant professor )

Program (20 mins each: 15 mins presentation and 5 mins discussion)

  1. "Considerations about the application on MANET"
    島田 秀輝 ( インターネット・アーキテクチャ講座 : COE特任助手 )
    Hideki Shimada ( Internet Architecture and Systems Lab. : COE assistant professor )

    [Abstract]
    In the ubiquitous network environment, MANET (Mobile Ad-hoc Network) is one of the technology communicating with other computers. MANET is the wireless multi-hop network that mobile computers configure autonomously without network infrastructures. In the area of MANET, there are many research topics of routing protocols and MAC protocols. Moreover, most topics about MANET are theoretical and are evaluated in the simulator not experiments in the real spaces. However, there are few practical applications to use daily life. We aim at construction of applications for daily use on MANET. We have developed the location-aware information sharing system among MANET nodes. In this presentation, we show this developed system that is demonstrated at the COE festival.

    Top
  2. "View Planning by a Data Acquisition Rate Map from Voxel Model which has Object Existence Probability"
    浅井 俊弘 ( 視覚情報メディア講座 : D3 )
    Toshihiro Asai ( Vision and Media Computing Lab. : D3 )

    [Abstract]
    Recently, various 3D modeling methods using a laser rangefinder are widely investigated. However, the shape data acquisition of wide area outdoor environments requires high cost. We propose a view planning method which presents a sensor operator a 2D map in which data acquisition rates are shown, in the case of generating a 3D model of outdoor environments by using a laser rangefinder. In our proposed method, voxel model is used as 3D shape model to compute acquisition data rate, because reconstruction is simple. 3D shapes are displayed with the objects existence probability inside of voxel. Experiments on outdoor environments have shown the feasibility of the proposed method.

    Top
  3. "Extracting Clinical Trial Design Information from MEDLINE Abstracts"
    原 一夫 ( 自然言語処理学講座 : D3 )
    Kazuo Hara ( Computational Linguistics Lab. : D3 )

    [Abstract]
    Evidence-based medicine (EBM) requires medical practitioners to select appropriate treatments for individual patients based on the current best evidence, and the results of phase III clinical trials are the major source of such evidence. In this presentation, we report results of experiment in extracting important information for EBM from the abstracts of phase III clinical trials, in an effort to investigate how far the existing natural language processing (NLP) techniques could support EBM using MEDLINE database.

    Top
  4. "Extracting aspect-aspect pairs for opinion extraction"
    小林 のぞみ ( 自然言語処理学講座 : D3 )
    Nozomi Kobayashi ( Computational Linguistics Lab. : D3 )

    [Abstract]
    Automatic extraction of human opinions from Web documents is a task has received increasing interest. In this work, we define an opinion as a chain consisting of the subject being evaluated, the aspect in which it is evaluated, and that evaluation. Based on this definition, we divide opinion extraction into two subtasks: extraction of pairs, and extraction of pairs. In this presentation, we focus on the extraction task using machine learning. We report on the results of the experiments for evaluating the results of extraction, and discuss future directions for this work.

    ==================== Break (10 min) ====================


  5. "A Device Management System Using Access Right Delegation for Peripheral Sharing Technologies Based on Virtual Bus Drivers"
    広渕 崇宏 ( インターネット・アーキテクチャ講座 : D3 )
    Takahiro Hirofuchi ( Internet Architecture and Systems Lab. : D3 )

    [Abstract]
    We are working on an advanced device sharing technology, in which virtual bus drivers transfer I/O requests of standardized peripheral interfaces over IP networks. Computers can share their peripheral devices with their full functionality between different operating systems; virtually-attached devices are used by unmodified drivers and applications in the same way as physically-attached ones. The I/O performance of virtually-attached devices is sufficient for practical usages in local area networks. However, device management systems on existing operating systems were basically designed for only physically-attached devices, which cannot provide dynamic access control against remote devices and flexible device usability for users. In this presentation, we propose a novel device management model based on access right delegation, and design a feasible implementation for existing operating systems by exploiting their hot-plug functions developed for recent peripheral interfaces. Our prototype implementation has achieved fine-grained access control by delegating access rights dynamically and users can manage both physically- and virtually-attached devices in an integrated and straightforward manner.

    Top
  6. "Design and Evaluation of Items Recommendation mechanism on a distributed agents environment"
    新井 イスマイル ( インターネット・アーキテクチャ講座 : D3 )
    Ismail Arai ( Internet Architecture and Systems Lab. : D3 )

    [Abstract]

    In ubiquitous environment, a recommendation method reflecting user's circumstance to acquire information rapidly and efficiently is expected. Also, it is desired that the recommendation system is scalable for a huge amount of information in the future. To solve this problem, we proposed a recommendation mechanism for distributed agents environment. The recommendation mechanism compares user profiles with contents metadata to find out suitable item for users. And these functions are implemented as agents on the P2P network based on location information. I present about availability of the proposed method through the experiment in a shopping mall.
    Top
  7. "Recognition of In-Hand Manipulation along with Rolling Contact using Orbital Motion of Contact Points on Object Surface"
    近藤 誠宏 ( ロボティクス講座 : D3 )
    Masahiro Kondo ( Robotics Lab. : D3 )

    [Abstract]
    We are developing a direct teaching system for a multifingered robot hand. In this system, teaches instinctively by manipulating an object attached sensor. And the multifingered robot hand realizes a manipulation similar to the input manipulation. In this paper, we propose a recognition method of object manipulation using contact state transition on the object surface. Pressure distribution sensor is directly attached to the manipulated object in this method. The manipulation is recognized by the orbital motion of contact points generated by rolling contacts. In order to evaluate the proposed method, a recognition experiment is conducted. It shows the effectiveness of the method and the importance of the contact information. Target tasks are In-hand manipulation which can change the position/orientation of the object using only finger's motion.

    Top
  8. "GPU-based Active Camera Assignment Scheme for High Resolution 3D Video"
    Sofiane Yous ( Artificial Intelligence Lab. : D3 )

    [Abstract]
    In my last presentation, I have presented a method for assigning multiple cameras to appropriate parts of a moving object, mainly an acting human body, for high fidelity 3D video. The camera system was made up of active Pan-Tilt (PT) cameras with long focal length lenses for high resolution input images. However, such cameras can capture only partial views of the object. The goal was to assign the camera set to the different parts of the moving object so as to achieve a high visibility of the whole object. The scheme consisted in evaluating the visibility to the different regions of the object, corresponding to different camera orientations and with respect to its field of view (FOV). Thereafter, each camera is assigned to one orientation in such a way to maximize the visibility to the whole object. The evaluation of this scheme have shown a good performance in terms of visibility. The remaining issue to be addressed is how to accelerate the processing time in order to allow our scheme to be useful in real systems. In this presentation, I will present an accelerated implementation based on General Purpose GPU techniques. The GPU, which is no longer useful just for graphics, can be used as a massive parallel stream co-processor. Thanks to its flexible programmability and huge memory bandwidth, it can increase dramatically the performance of our assignment scheme.

    Top

21st Century COE Program
NAIST Graduate School of Information Science