NAIST Graduate School of Information Science
| Date: | Tuesday, Jul. 19, 2005 |
|---|---|
| Time: | 13:30 - 16:20 |
| Place: | L2 Lecture Room |
| Language: | English (Oral Presentation), English/Japanese (Question) |
| Chairperson: | Muneyuki Sakata (Image Processing Lab. : PD), Osamu Tobe (Internet Architecture and Systems Lab. : D3) |
Nowadays many kinds of device become widely used to obtain location information. Howerver each device has its own interface and notation of location information, and these devices are operated individually. This makes it quite difficult to use multiple devices in a system, and gives inconvenience to users who want to use the services using location information.
In this study I propose the model to deal with location information from several devices uniformly. This model consists of two separated layers: positioning layer and uniformed location information layer. By adopting this model, users are able to use location information without any care for the type of devices from which it is obtained.
There is a growing interest in developing applications that are attentive to user's needs. This kind of applications is often referred to as Attentive or NonCommand Interfaces. These applications exploit several communication channels with a specific aim to identify and follow the user's intent. Usually, this is done by estimating the user's attention. Having this information, the application can then propose a service according to the user's needs at the moment. From this point of view, we propose to use gaze information as an indicator of user's attention. Blended with other communication channels or taken alone, the gaze has a complicated behavior and can be very difficult to interpret. We propose an application that uses gaze information to detect the degree of understanding of non-native users reading an English text. Being able to discover when a user is having understanding troubles, the application proposes a translation of the word(s) that is(are) causing the trouble. An experimental study of gaze behavior during reading tasks is presented and characteristic features are identified.
A direct teaching method is effective for multifingerd robot hand designed similar to human's hand structure. To achieve an instinctive teaching, it is necessary to recognize manipulation patterns by observing a human hand. In this paper, a recognition system of object manipulation is developed as a part of direct teaching system for multifingered robot hand. Conventional recognition systems have difficulties in recognizing complex manipulations obtained by dextrous finger actions due to the low accuracy of the data glove. This paper proposes an operation measurement system using an object that is specially designed for teaching. This object consists of a six-degree-of-freedom position/orientation sensor and a tactile sensor sheet. This paper also proposes a method of spotting recognition by observing a change of the contact state of the palm surface. The effectiveness of this method is confirmed by experiments.
We propose a non-scan design-for-testability (DFT) method that guarantees complete fault efficiency (FE) for register transfer level (RTL) data paths. We first define the partially strong testability as a characteristic of data paths.
Then we propose a DFT method to make a data path partially strongly testable and a test generation method for partially strong testable data paths based on the time expansion model (TEM).
The proposed DFT method can reduce hardware overhead drastically compared with the previous method based on strong testability. Moreover, the proposed DFT method can generate test patterns with complete FE in practical time and allow at-speed test.
VLSI technology has been growing, accordingly the speed of VLSI circuits has accelerated in recent years. Delay testing is an important technology to guarantee the timing correctness for a high speed circuit. In the semiconductor industry, scan design is widely accepted as a design for testability (DFT) method to reduce the cost for testing. However, this approach has some drawbacks: high hardware overhead and long test application time and so on. This presentation introduces a new concept of hierarchical testability called Single-Port-Change (SPC) two-pattern testability. We propose a non-scan DFT method which makes each path that needs to be tested in a data path SPC two-pattern testable.
Many systems around us can be considered as hybrid dynamical systems, which has continuous and discrete variables. In chemical processing operations, some dynamical behaviour also can be described as hybrid dynamical systems. In this presentation, we apply MPC (Model Predictive Control) with multiple models to a highly nonlinear chemical reactor that has mode changes. We show that a MPC with multiple models has good performance capabilities in a control without the foreknowledge of mode changes.
In the extraction and recognization of temproal information in Chinese, syntactic analysis is necessary. Because the temporal relation between verb and temporal phrase are ambiguious. We will adopt the dependency structure for extracting the relations. In this presentation, We present a method for constracting dependency structure analysis of Chinese. Our bottom-up deterministic analyzer arrange from original Nivre’s algorithm (Nivre and Scholz, 2004). Support Vector Machines (SVMs) are utilized to determine the word dependency relations. We adopt two methods to improve the analyzer. Dividing the sentence by root node and using prepositional chnker. This bottom-up analyzer doesn’t use top-down information. We supply the top-down information by constructing SVMs based root node finder to solve this problem. The analyzer cannot analyze the long PP correctly. We used a SVMs based chunker to extract the PP before the dependency analysis. Experimental evaluation on the Penn Chinese Treebank Corpus shows that the proposed extensions improve the parsing accuracy significantly.
In this presentation, I propose a new parsing framework based on a unification grammar. This framework is notable in that it is constructed of two new components, a failure pool and an error recovery module. Most of the past systems cannot process the input that is ill-formed or outside of the coverage of their grammars. In contrast, this framework stores all unification failures during the parsing process, and if the system fails to parse the input, it searches for a failure with the least severe violations in the failure pool, and it tries to recover from the failure in order to continue the parsing process. I present the details of this framework and also report on the current progress and the future direction of this research.