Search

Entity alignment (EA) aims to find equivalent entities in differentknowledge graphs (KGs). State-of-the-art EA approaches generallyuse Graph Neural Networks (GNNs) to encode entities. However,most of them train the models and evaluate the results in a full-batch fashion, which prohibits EA from being scalable on large-scaledatasets. To enhance the usability of GNN-based EA models in real-world applications, we present SEA, a scalable entity alignmentsystem that enables to (i) train large-scale GNNs for EA, (ii) speedup the normalization and the evaluation process, and (iii) reportclear results for users to estimate different models and parametersettings. SEA can be run on a computer with merely one graphiccard. Moreover, SEA encompasses six state-of-the-art EA modelsand provides access for users to quickly establish and evaluate theirown models. Thus, SEA allows users to perform EA without beinginvolved in tedious implementations, such as negative samplingand GPU-accelerated evaluation. With SEA, users can gain a clearview of the model performance. In the demonstration, we show thatSEA is user-friendly and is of high scalability even on computerswith limited computational resources.

Based on electromagnetic theory, metallurgical characteristics can be detected by eddy current nondestructive testing technology. In this study, the relationship between the surface microstructure and the eddy current output of martensitic stainless steel AISI 410 was studied using this technology at different quenching temperatures. The mechanical properties include material hardness, microstructure types and microstructural changes after thermal treatment was evaluated. Using Vickers hardness as the surface hardness index of AISI 410 steel, the relationship between eddy current output signal, in terms of impedance and inductance, and sample surface hardness was studied and the effects of different quenching temperatures on the steel&rsquo, s surface hardness was examined. In addition, the change of microstructure types of AISI 410 steel after thermal treatment was detected by the eddy current nondestructive testing method, and the results were verified by metallographic microscopy.

The length of transition section of Metro trapezoidal sleeper track bed affects the overall stiffness of the track. This paper establishes the Metro vehicle-rail coupling model and compares the calculated results of the trapezoidal sleepers of different transition lengths with the field measured data to verify the correctness of the model. The calculation results show that: The maximum vertical displacement of the track bed at the length of 0, 3, 6, and 9 meters is 2.55, 2.53, 2.01, and 1.92 mm, respectively, the maximum vertical displacements of the transition rails at the length of 0, 3, 6, and 9 meters are 2.53, 2.44, 2.31, 2.30 mm, respectively, the maximum rotation angles of the 0, 3, 6, and 9 meters transition sections are 1.01 rad ‰, 0.96 rad ‰, 0.90 rad ‰, 0.84 rad ‰, respectively. Considering the safety of the track and other factors, the length of the trapezoidal sleeper transition section should not be less than 9m.

doctoral, 創造科学技術大学院, 甲第848号

The side-lobe level is an important factor that determines the performance of the Multibeam Echo Sounders. In order to keep the side-lobes down, the magnitude and phase discrepancies must be compensated prior to the beamforming operation. In this paper, a calibration method of MBES's received array is proposed. This method can be implemented in room other than an open water area., and it doesn't require standard hydrophones or the high precise measurement-equipment. In the paper, a received array of MBES, whose frequency is 200 kHz, is calibrated, and the results have shown that the decreased of the side-lobe level could be as low as 9-12 dB.

This paper discusses a technique for the classification of divers, underwater vehicles and other similar targets based on adaptive multi-feature fusion. This technique mainly uses target features, e.g., image feature, moment feature and HRRP (high resolution range profile) features selected by an Adaptive Feature Fusion method. The sea trial data collected in Sanya in January, 2009 are processed by the algorithm. Real data processing results demonstrate the Adaptive Multi-Feature Fusion based technique can effectively reduce the error probability.