The 3rd COE Postdoctoral and Doctoral Researchers
Technical Presentation

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

1. "3D Modeling of Outdoor Scenes by Integrating Stop-and-Go and Continuous Scanning of Rangefinder "
   Toshihiro Asai (Vision and Media Computing Laboratory : D2)
   浅井 俊弘 (視覚情報メディア講座 : D2)
   

   [Abstract]
   We propose a 3D modeling method for wide outdoor environments by integrating stop-and-go and continuous scanning of laser rangefinder. First, range images of an outdoor scene are measured by stop-and-go scanning by using omnidirectional laser rangefinder, and the 3D surface model is generated. Next, to recover the parts that are not measured in the stop-and-go scanning mode due to occlusion, we measure the outdoor scene in the continuous scanning mode where the laser rangefinder line-scans the scene under movement is employed. In the continuous scanning mode, the position and orientation of the rangefinder is acquired by a hybrid sensor which consists of GPS (Global Positioning System) and INS (Inertial Navigation System). Finally, multiple range data acquired in two scanning modes are integrated by registering overlapped parts of range data.
   
   Top

2. "Support the Design of User Interfaces for Plant Operations"
   Xiwei Liu (Systems and Control Laboratory : D2)
   劉 希未 (システム制御・管理講座 : D2)

   [Abstract]
   User interfaces for plant operations are safety-critical. In order to support the design of effective user panels for a supervisory control system, we propose a human perception model to evaluate and improve the panels from the viewpoint of perception. Based on psychological and psychophysical researches, the model interprets visual field, eye movement, and visual performance. After extracting the information of graphic items and process variables from a user panel, we employ the perception model to scan over the panel. According to the scanning results, visual strength and density checks are processed. In light of some weak points and their causes we found, the panel is modified and reevaluated. This evaluation procedure is an iterative course until there is no weak point on the panel. We validate the usefulness of this approach by several case studies.
   
   Top

3. "Evaluation of SIMO-model-based ICA with Self-Generator for Initial Filter for Blind Source Separation"
   Tomoya Takantani (Speech and Acoustics Laboratory : D2)
   高谷　智哉 (音情報処理学講座 : D2)

   [Abstract]
   We address the blind decomposition problem of the binaural mixed audio signals to realize the audio augmented reality system. In order to solve the problem, we have proposed Single-Input Multiple-Output (SIMO)-model-based indepednet component analysis (ICA) (SIMO-ICA) algorithm which decompose the mixed observed signals into the SIMO-model-based signals. Here the term "SIMO" represents the specific transmission system in which the input is a single source signal and the outputs are its transmitted signals observed at multiple microphones. In our previous work, in order to improve the decomposition performance, we have proposed a novel blind separation method using SIMO-ICA with a self-generator (SG) for the initial filter. Although an attractive feature of SIMO-ICA is beneficial to the binaural sound separation, SIMO-ICA has a serious drawback in its high sensitivity to the initial settings of the separation filter. In the proposed method, the SG functions as thepreprocessor of SIMO-ICA, and it can provide a valid initial filter for SIMO-ICA. In this presentation, we discuss the spatial quality of the output signals in the proposed and conventional methods. To evaluate it, opinion tests are performed. The experimental results reveal that the separation performance of the proposed method is superior to those of conventional methods.
   
   Top

4. "USB/IP ― A Peripheral Bus Extension for Device Sharing over IP Network"
   Takahiro Hirofuchi (Internet Architecture and Systems : D2)
   広渕　崇宏 (インターネットアーキテクチャ講座 : D2)

   [Abstract]
   In this presentation, we propose USB/IP as a peripheral bus extension over an Internet Protocol (IP) network. This novel device sharing approach is based on the sophisticated peripheral interfaces that are supported in most modern operating systems. Using a virtual peripheral bus driver, users can share a diverse range of devices over networks without any modification in existing operating systems and applications. Our experiments show that USB/IP has sufficient I/O performance for many USB devices, including isochronous ones. We also describe performance optimization criteria that can be used to achieve further performance improvements.
   
   Top

5. "Query generation mechanism for the information retrieval reflecting users' circumstance "
   Ismail Arai (Internet Architecture and Systems : D2)
   新井 イスマイル (インターネットアーキテクチャ講座 : D2)

   [Abstract]
   To search the contents through WWW in consideration of user's circumstance condition, search engines should make use of some information such as current time, user's location, user's schedule and so on. We propose an information retrieval method that effectively uses both user's and content's metadata based on TPO (Time, Position, Occasion). We implement a prototype of information retrieval system based on the proposed method, which consists of query generation part, contents matching part, and matched results scoring part. In our query generation part, we reduce user's overload of choosing metadata. And a user becomes possible to retry sorting the matching results.
   
   Top

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

6. "Error Analysis of Noun Phrase Anaphora Resolution using a Machine Learning-based Approach"
   Ryu Iida (Computational Linguistics Laboratory : D2)
   飯田 龍 (自然言語処理学講座 : D2)

   [Abstract]
   We proposed an approach to anaphora resolution where the model first identifies the most likely antecedent candidate and then determines an anaphoricity for a given anaphor candidate. Next, we will try to analyze manually errors of Japanese noun phrases anaphora resolution. In this talk, we will report the results of recently conducted error analysis and discuss future direction of my research.
   
   Top

7. "Computing Citation Relatedness Using Kernels"
   Takahiko Ito (Computational Linguistics Laboratory : D2)
   伊藤 敬彦 (自然言語処理学講座 : D2)

   [Abstract]
   We apply a family of graph kernels, to the computaton of relatedness between technical documents based on their citation information. The advantage of using these kernels is that the diffusion process underlying its computation allows them to capture the relation between documents even when these documents do not cite or are not cited by the same documents. We compare the performance of these kernels with that of traditional co-citation and co-reference method, using real data.
   
   Top

8. "Problems associated with applying NLP techniques to Clinical Trial MEDLINE Abstracts"
   Kazuo Hara (Computational Linguistics Laboratory : D2)
   原 一夫 (自然言語処理学講座 : D2)

   [Abstract]
   Acquiring up-to-date information from the results of clinical trials is essential for practitioners of Evidence-Based Medicine in the medical community. However, the fact is this information is likely to be buried among the large amount of published articles or even among MEDLINE, the most popular database of medical paper abstracts. Our research goal is to provide an easy way for medical doctors or patients to access this information, which includes indexing clinical trial MEDLINE abstracts according to the clinical trial phase or subject patient population, in addition to extracting key information with respect to the clinical trial design such as compared treatments in the trial. Furthermore, there are problems associated with applying conventional natural language processing (NLP) techniques to information extraction from clinical trial MEDLINE abstracts, one such problem caused by the frequent appearance of coordinate structures and technical terms in the abstracts. In this presentation, we will give an overview of our plan to solve the problems.
   
   Top

9. "Opinion mining as extraction of attribute-value pairs"
   Nozomi Kobayashi (Computational Linguistics Laboratory : D2)
   小林 のぞみ (自然言語処理学講座 : D2)

   [Abstract]
   This paper proposes a method of extracting opinions appearing in freely written texts such as on message boards or weblog pages on the Web. Assuming that an opinion can be represented as a tuple , we propose a machine learning based method to extract such tuples from texts. In this method, the major task is decomposed into (a) the process of extracting attribute-value pairs from given texts and (b) the process of judging whether an extracted pair forms an opinion of the author. To both of the tasks we apply machine-learning techniques. We also report on the present results of our experiments and discuss future directions.
   
   Top

NAIST Graduate School of Information Science