NEC

About: NEC is a company organization based out in Tokyo, Japan. It is known for research contribution in the topics: Signal & Layer (electronics). The organization has 33269 authors who have published 57670 publications receiving 835952 citations. The organization is also known as: NEC Corporation & NEC Electronics Corporation.

Distributed multiparty desktop conferencing system: MERMAID

Abstract: This describes a distributed multiparty desktop conferencing system (MERMAID) and presents its preliminary brief evaluation, obtained as a result of daily use. MERMAID, which is designed based on group collaboration system architecture, provides an environment for widely distributed participants, seated at their desks, to hold real-time conferences by interchanging information through video, voice, and multimedia documents. This system is implemented by using narrow-band ISDN, high-speed data network, and UNIX-based EWSs with electronic writing pads, image scanners, video cameras, microphone-installed loudspeakers, etc. The system provides participants with the means for sharing information in such multimedia forms as video images, voice, text, graphics, still images, and hand drawn figures.

High-performance TFTs fabricated by XeCl excimer laser annealing of hydrogenated amorphous-silicon film

Abstract: High-performance staggered a-Si:H and poly-Si thin-film transistors (TFTs) fabricated by XeCl excimer laser annealing of a-Si:H films are discussed. The field-effect mobility of poly-Si TFT is 102 cm/sup 2//V-s, and that of a-Si:H TFT is 0.23 cm/sup 2//V-s. Their drain current on/off ratios are over 10/sup 6/. Except for the crystallization, the fabrication process was the same for both of them. This process appears extremely promising for the integration of matrix elements and peripheral drivers in a single substrate. >

Boosting in the limit: maximizing the margin of learned ensembles

Abstract: The "minimum margin" of an ensemble classifier on a given training set is, roughly speaking, the smallest vote it gives to any correct training label. Recent work has shown that the Adaboost algorithm is particularly effective at producing ensembles with large minimum margins, and theory suggests that this may account for its success at reducing generalization error. We note, however, that the problem of finding good margins is closely related to linear programming, and we use this connection to derive and test new "LPboosting" algorithms that achieve better minimum margins than Adaboost.However, these algorithms do not always yield better generalization performance. In fact, more often the opposite is true. We report on a series of controlled experiments which show that no simple version of the minimum-margin story can be complete. We conclude that the crucial question as to why boosting works so well in practice, and how to further improve upon it, remains mostly open.Some of our experiments are interesting for another reason: we show that Adaboost sometimes does overfit--eventually. This may take a very long time to occur, however, which is perhaps why this phenomenon has gone largely unnoticed.

Element-Selective Single Atom Imaging

Abstract: Electron energy-loss spectroscopy (EELS) is widely used to identify elemental compositions of materials studied by microscopy. We demonstrate that the sensitivity and spatial resolution of EELS can be extended to the single-atom limit. A chemical map for gadolinium (Gd) clearly reveals the distribution of Gd atoms inside a single chain of metallofullerene molecules (Gd@C82) generated within a single-wall carbon nanotube. This characterization technique thus provides the "eyes" to see and identify individual atoms in nanostructures. It is likely to find broad application in nanoscale science and technology research.

MDL-based context-dependent subword modeling for speech recognition

Abstract: Context-dependent phone units, such as triphones, have recently come to be used to model subword units in speech recognition systems that are based on the use of hidden Markov models(HMMs).While most such systems employ clustering of the HMM parameters(e.g., subword clustering and state clustering)to control the HMM size, so as to avoid poor recognition accuracy due to a lack of training data, none of them provide any effective criteria for determining the optimal number of clusters.This paper proposes a method in which state clustering is accomplished by way of phonetic decision trees and in which the minimum description length(MDL)criterion is used to optimize the number of clusters.Large-vocabulary Japanese-language recognition experiments show that this method achieves higher accuracy than the maximum-likelihood approach.