The 10th COE Postdoctoral and Doctoral Researchers
Technical Presentation

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

1. "Matrices of Multiple Weights for Test Response Compaction with Unknown Values"
   Thomas Clouqueur (Computer Design and Test Laboratory : PD)
   トマ クルッカー (コンピュータ設計学講座 : PD)
   

   [Abstract]
   This presentation focuses on the compaction of the outputs of VLSI circuits during test. First, we will see why such compaction is needed during testing of chips and what methods were previously developed for that compaction. Then, we will see how the occurrence of unknown states/values in the response, a common phenomenon, impacts the compaction. We will propose new matrices for linear test output compactors that offer superior compaction performance in the presence of unknown values.
   
   Top

2. "Evaluation of a Positioning Method Using IrDA Sensors and a Pedometer"
   Ryuhei Tenmoku (Vision and Media Computing Laboratory : D2)
   天目 隆平 (視覚情報メディア講座 : D2)
   

   [Abstract]
   Applying augmented reality techniques for wearable computers, many applications can be realized. We have proposed a wearable augmented reality system which can present location-based information to the user by annotating real scene image. Our system measures the user's position using IrDA sensors and a pedometer. The main topic of this presentation is experiments for evaluating this positioning method. This presentation also describes information presentation techniques for wearable augmented reality systems which is my current work.
   
   Top

3. "A Category-oriented Approach to Paraphrase Corpus Construction"
   Atsushi Fujita (Computational Linguistics : D3)
   藤田 篤 (自然言語処理学講座 : D3)
   

   [Abstract]
   Given an increasing number of digital documents available on computer networks, we often find that multiple documents report the same information using different wording, i.e., paraphrases. To realize intelligent technologies for content processing, handling paraphrases is an indispensable issue. In this presentation, we introduce our ongoing work on accumulating paraphrase examples. The resulting corpus can be used for the following tasks: evaluation of paraphrase generation systems, induction of paraphrase patterns, and identification of linguistic knowledge useful for paraphrase generation. We propose an effective scheme which has the following features. First, it employs an existing paraphrase generation system. The human labor is required only to judge the appropriateness of each automatically generated paraphrase. Second, it generates paraphrases based on given category of paraphrasing rules. This category-oriented approach allows us to exhaustively retrieve paraphrasable sentences from a given text collection. We have constructed two corpora each of which contains one thousand paraphrases. One is comprised of "paraphrases of light-verb construction," and the other of "transitivity alternation." The advantage of this approach over conventional ones is the reduction in work-hours required.
   
   Top

4. "IBAS : Internet-Based Augmentation System for GNSS"
   Osamu Tobe (Information Technology Center : D2)
   戸邉 論 (情報科学センター : D2)
   

   [Abstract]
   In mobile and ubiquitous computing, it is important to locate nodes in high accuracy. Global Navigation Satellite Systems (GNSS), such as Global Positioning System (GPS) is one of the effective devices. In order to enhance the position accuracy of GNSS, it is necessary to send "error correction data" to these nodes. In this presentation, I propose the Internet-Based Augmentation System for GNSS (IBAS) which provides error correction data with mobile nodes over the Internet. Then, I also refer to how to distribute location-based information such as error correction data.
   
   Top

5. "Immersive Telepresence System Using High-resolution Omnidirectional Video with Locomotion Interface"
   Sei Ikeda (Vision and Media Computing Laboratory : D2)
   池田 聖 (視覚情報メディア講座 : D2)
   

   [Abstract]
   We propose a novel telepresence system and a method to generate virtualized environments from real images. The system consists of a multi-projection display and a treadmill. It is designed to enable a user to move his view point freely by his actual walking in a photorealistic virtualized environment. For this system, the virtualized environment is generated as video streams by using computer vision techniques from real images acquired by an omnidirectional multi-camera system (OMS). In this method, shake effect and replay speed of video are corrected to improve the sense of walking. For these corrections, calibration and motion estimation of the OMS are performed in advance. The system provides users with rich sense of walking in a remote site.
   
   Top

NAIST Graduate School of Information Science