Mitsubishi Electric

About: Mitsubishi Electric is a company organization based out in Ratingen, Germany. It is known for research contribution in the topics: Signal & Voltage. The organization has 23024 authors who have published 27591 publications receiving 255671 citations. The organization is also known as: Mitsubishi Electric Corporation & Mitsubishi Denki K.K..

System for correcting improper determiners

Abstract: In a grammar checking system which includes first tagging a sentence as to parts of speech, an improper determiner correction module detects improper referents of a noun phrase and suggests insertion of a determiner should one be necessary or the deletion of an improper determiner. To detect improper use of a determiner, parts of speech tags are utilized to characterize a sentence, thus to identify noun phrases by maximally matching a pattern that defines which sequences of parts of speech tags constitute valid noun phrases. This is accomplished by identifying the start of the noun phrase and its end to permit checking for either missing determiners, extraneous determiners, or lack of agreement in number for the constituents of the noun phrase. Once a noun phrase is found, the system tests to see if the noun is a head noun and thereafter if the head noun is a singular non-proper noun, whether the noun phrase contains a determiner, whether the noun phrase is a title, whether the head noun is a mass noun or whether the noun phrase is part of an idiom. The system also checks to see if the head noun is a proper noun and if so whether the noun phrase contains a determiner. Finally, if the noun phrase contains a determiner the system checks if the determiner agrees in number with the head noun. In this manner, both missing determiners, extraneous determiners and determiner noun number disagreement are detected and reported.

Double-lattice photonic-crystal resonators enabling high-brightness semiconductor lasers with symmetric narrow-divergence beams.

Abstract: Achieving high brightness (where brightness is defined as optical power per unit area per unit solid angle) in semiconductor lasers is important for various applications, including direct-laser processing and light detection and ranging for next-generation smart production and mobility. Although the brightness of semiconductor lasers has been increased by the use of edge-emitting-type resonators, their brightness is still one order of magnitude smaller than that of gas and solid-state/fibre lasers, and they often suffer from large beam divergence with strong asymmetry and astigmatism. Here, we develop a so-called ‘double-lattice photonic crystal’, where we superimpose two photonic lattice groups separated by one-quarter wavelength in the x and y directions. Using this resonator, an output power of 10 W with a very narrow-divergence-angle (<0.3°) symmetric surface-emitted beam is achieved from a circular emission area of 500 μm diameter under pulsed conditions, which corresponds to a brightness of over 300 MW cm−2 sr−1. In addition, an output power up to ~7 W is obtained under continuous-wave conditions. Detailed analyses on the double-lattice structure indicate that the resonators have the potential to realize a brightness of up to 10 GW cm−2 sr−1, suggesting that compact, affordable semiconductor lasers will be able to rival existing gas and fibre/disk lasers. An optimized design for a broad-area surface-emitting photonic-crystal laser leads to high brightness of over 300 MW cm–2 sr–1 and an output power of 10 W under pulsed excitation.

Joint CTC/attention decoding for end-to-end speech recognition

Abstract: End-to-end automatic speech recognition (ASR) has become a popular alternative to conventional DNN/HMM systems because it avoids the need for linguistic resources such as pronunciation dictionary, tokenization, and context-dependency trees, leading to a greatly simplified model-building process. There are two major types of end-to-end architectures for ASR: attention-based methods use an attention mechanism to perform alignment between acoustic frames and recognized symbols, and connectionist temporal classification (CTC), uses Markov assumptions to efficiently solve sequential problems by dynamic programming. This paper proposes joint decoding algorithm for end-to-end ASR with a hybrid CTC/attention architecture, which effectively utilizes both advantages in decoding. We have applied the proposed method to two ASR benchmarks (spontaneous Japanese and Mandarin Chinese), and showing the comparable performance to conventional state-of-the-art DNN/HMM ASR systems without linguistic resources.

Sandwich-type ferromagnetic RF integrated inductor

Abstract: The first demonstration of a sandwich-type ferromagnetic RF integrated spiral inductor for the 2-GHz range is reported. Two ferromagnetic CoNbZr films were set to sandwich the spiral in order to enhance the amount of magnetic flux linkage across the coil current. The stresses given from the insulator to the ferromagnetic film were studied. The inductance L of 7.9 nH and the qualify factor Q of 12.7 were obtained for a 200 /spl mu/m/spl times/400 /spl mu/m size four-turn rectangular spiral at f=2 GHz. The inductance was better than that of an air core of the same coil size by 19%, and the Q was better by 23%. Comparison with the on-top magnetic film type was also discussed.

Image processing device and image processing method

Abstract: The invention relates to an image processing device and an image processing method an image based on retrieved data can be displayed before authentication of the retrieved data is completed While the authentication unit(104) is performing authentication, the resource access supervision section(105) disallows access to the local resource and sends the display image generating unit(108) an authentication-in-progress notification Upon receiving this notification, the display image generating unit(108) generates a notification image indicating that authentication is being performed, and generates the display image by combining the second partial images and the notification image the invention enables an image based on retrieved data to be displayed before authentication of the retrieved data is completed, without compromising the security of hardware resources When authentication succeeds, the display image generating unit may access the local resource(107) and generate the display image by combining the first and second partial images