Hosei University Repository
Not a member yet
22008 research outputs found
Sort by
SPEEDUP OF FLUX WAVEFORM CONTROL BASED ON DEEP NEURAL NETWORK IN SINGLE SHEET TESTER
Get PDFTo achieve high efficiency of various motors, the realization of the high saturation magnetic flux density and thinning of the electrical steel sheets are required. When the performances of various motors are verified using the finite element method (FEM), the accurate magnetic property of the electrical steel sheet is applied to FEM. However, because extended elapsed time is taken to measure the magnetic property in the high flux region when the conventional feedback control is applied to the flux waveform control, the method for reducing the measurement time is essential. In this paper, a waveform control method that evaluates the initial waveform using a deep neural network (DNN) is proposed. The performance of the proposed system supported by the DNN is verified and compared to that supported by only the conventional feedback formula
USING GENETIC ALGORITHMS AND NEURAL NETWORKS AUTOMATIC DERIVATION OF STRASSEN’S MATRIX PRODUCT ALGORITHM
Get PDFStrassen\u27s matrix product algorithm is a fast method for computing matrix products. By using the main principles of this algorithm, the computation of the nth-order matrix product can be significantly reduced.The purpose of this study is to confirm whether Strassen\u27s matrix product algorithm can be automatically derived using Genetic Algorithm and Neural Network. Specifically, we use genetic algorithms to determine the structure of the neural network, including interconnections. Although we could not obtain very practical results for the degree of adaptation, we considered that this was due to the fact that the combination of addition, subtraction, and multiplication was left to the genetic algorithm, which resulted in a different combination of calculations from those that were originally necessary. We should consider the composition of neural networks and increase the number of constraints
GRAPH STRUCTURING FOR CORPORATE BUSINESS DOCUMENTS AND FEATURE WORD EXTRACTION FROM THE GRAPH USING GRAPH CONVOLUTIONAL NETWORKS
Get PDFThis study proposes a graph structuring and feature word extraction method for business information documents extracted from corporate websites for corporate and market trend research. The graphs created from business documents consist of a hierarchical structure graph reflecting the relationship between businesses and companies, and a dependency graph using the dependency relations of business documents. For feature word extraction, RGCNs were used for the dependency graph, and the F value was 0.33 for the business document graph consisting of a single company, and 0.37 for the business document graph consisting of multiple companies in a single industry
GENERATION OF ZnO MICRO PARTICLES IN GAS AND LIQUID PHASE ENVIIROMENTS BY LASER ABLATION USING A PRESSURE VARIABLE CELL
Get PDFWe study the structure and properties of atoms, molecules, and microparticles in low temperature environments such as superfluid helium and liquid nitrogen. To further discuss laser spectroscopy of alkali atoms in a superfluid helium environment and the generation of micro particles by laser ablation, we have been developing a pressure variable cell that can vary pressure independent of temperature. In particular we have improved a pressure variable cell that can independently change the pressure in the gas and liquid phases. Also, we performed ZnO particle generation by laser ablation using this pressure variable cell when the pressure, ambient gas, and temperature were varied. As a result, the effects of the above changes on the shape and size of the ZnO particles are discussed
EVALUATION OF EXPECTED UTILITY AND ASSET VALUE WITH RESPECT TO OPTIMAL EXECUTION STRATEGY USING MEAN-FIELD GAME
Get PDFWe consider optimal strategies to maximize expected profits for traders. In our market, we assume 3 types of traders: the first one is a major trader whose purpose is liquidation of shares, the second one is minor traders whose purpose is only acquisition of profits, and the last one is noise traders who have no purpose. In derivation of optimal strategies of the major trader and minor traders, we use a mean-field game framework in order to grasp overall volumes of orders from minor traders based on [3]. In our numerical experiments, we implement each strategies using parameter values estimated from Japanese stock market data, and evaluate their expected utilities and the asset value at the maturity
Comparison of Reimbursement Pricing Systems for Medical Devices in Japan and Other Countries
Get PDFIn this article, we discuss the current status of the reimbursement price system for medical devices in Germany, France, the UK, the US, and Australia, which are the reference countries for the “foreign price adjustment system” in determining insurance reimbursement prices, and also provide implications for Japan. While Germany, the UK, and the US generally have reimbursement systems where medical devices are reimbursed as part of lumpsum payment, in other countries, as with Japan’s specified insurance medical devices, individual reimbursement prices are set. In countries that primarily implement lumpsum payments, there is a system of adding a certain period of reimbursement price increase to compensate for the decrease in medical institutions’ revenue resulting from the price increase of improved products. In addition, there are other support systems for collecting clinical evidence, which serve as examples for future discussions in Japan to promote the appropriate introduction of new medical devices
<Book Reviews> Kim Kyo-Seong, Baek Seung-Ho, Seo Jeong-Hee, Lee Sophia Seung-Yoon, The Age of Basic Incom is Coming
Get PDF<Book Reviews> UMEZAKI Osamu, Career Development and Industrial Relations in Japan : Survey-Based Labor Economics
Get PDFIntroducing Collaborative Multiple-Student Single-Teacher for Online Learning
Get PDFKnowledge distillation is a popular method where a large trained network (teacher) is implemented to train a smaller network (student). To decrease the need for training a much larger network (teacher) for real time application, one student self-knowledge distillation was introduced as a solid technique for compressing neural networks especially for real time applications. However, most of the existing methods consider only one type of knowledge and apply one-student one-teacher learning strategy.This thesis presents a collaborative multiple-student single-teacher system (CMSST). The proposed approach is based on real time applications that contain temporal information, which play an important role in understanding. We designed a backbone old student network with target complexity for deployment, during training, once the old student provides high-quality soft labels to guide the hierarchical new student, it also offers the opportunity for the new student to make meaningful improvements based on the students’ revised feedback via the shared intermediate representations. Moreover, we introduced soft target label smoothing technique to the CMSST.Experimental results showed that the accuracy can be improved on newly developed teacher knowledge distillation by 1.5% on the UCF-101. Also the accuracy was improved by 1.15% compared to normal huge teacher knowledge distillation on CIFAR100 dataset
