Search

Current researches indicate that inductive bias (IB) can improve Vision Transformer (ViT) performance. However, they introduce a pyramid structure concurrently to counteract the incremental FLOPs and parameters caused by introducing IB. This structure destroys the unification of computer vision and natural language processing (NLP) and complicates the model. We study an NLP model called LSRA, which introduces IB with a pyramid-free structure. We analyze why it outperforms ViT, discovering that introducing IB increases the share of high-frequency data in each layer, giving "attention" to more information. As a result, the heads notice more diverse information, showing better performance. To further explore the potential of transformers, we propose EIT, which Efficiently introduces IB to ViT with a novel decreasing convolutional structure under a pyramid-free structure. EIT achieves competitive performance with the state-of-the-art (SOTA) methods on ImageNet-1K and achieves SOTA performance over the same scale models which have the pyramid-free structure., 7 pages, 10 figures

Non-terrestrial networks (NTNs) are expected to provide continuous and ubiquitous connectivity across the globe. We identify that deterministic routing is a promising technology to achieve high resource utilization and strict QoE guarantee in NTNs. However, due to the time-varying topology, dynamic bandwidth and long propagation, for establishing deterministic services, it is not only difficult in finding stable end-to-end (E2E) routing, but also complex for joint routing and scheduling. Towards this, we propose an optimal solution based on temporal graphs to enhance resource utilization and service capability of NTNs. First, an extended time-expanded graph (ETEG) with introduced virtual nodes and edges is constructed to uniformly model both heterogeneous resources and traffic requirements. Then, we recast the deterministic routing problem as an integer linear programming (ILP) problem, by linearizing the constraints containing variable-dependent summations, round-down operation and multiplication of nonlinear terms. With the solid upper bound by ILP-based scheduling, we further propose a low-complexity ETEG-based shortest time-featured path algorithm to jointly use transmission and storage resources to support dynamic queuing and cross-cycle transmission of packets, thereby determining the best outgoing link and forwarding cycle hop-by-hop. The optimality and complexity are analyzed, together with simulations to verify the superiority of proposed algorithm in traffic acceptance, E2E delay and running time.

The fatigue damage of rock is an important factor affecting the stability of rock structure. In this paper, the mechanical response of coal under cyclic loading was studied. In order to accurately describe the deformation characteristics of coal under cyclic loading, an elastic-plastic model of coal based on the theory of subloading surface was established and verified by experiments. The model can well reflect the Mancin effect and ratcheting effect of coal samples, which is basically consistent with the actual deformation characteristics of coal, and the theoretical value and experimental value are in good agreement. At the same time, the cyclic response characteristics of specimens under strain load disturbance were analyzed. The results show that the specific strain disturbance can only cause a certain damage to coal and the area of hysteresis loop decreases first, then stabilizes, and then increases as the number of cycles increases. In addition, the damage factor Dn in the model was analyzed in this paper. Dn, which can accurately describe the damage process of coal, accurately locate the time point of disturbance load change, and has greater sensitivity to coal failure, is helpful to improve the accuracy of the stability judgment of coal structure and ensure the safety of engineering. The above results are of great significance for strengthening the understanding of coal mass instability process and mode under cyclic loading.

Deep-circling thermal water upwelling and trickling to high geo-temperature roadway obviously alter the microclimate in mines, which brings difficulty to the prediction of airflow temperature and humidity. This is the basis of air-conditioning cooling load calculation. The heat and mass transfer between trickling water and airflow is rather complicated. Moreover, humid air exhibits the accumulation effect of heat and humidity in the long-distance flow process. In this paper, an apparatus was designed and developed to explore the influence of thermal water trickling on the airflow thermal parameters of a section of roadway (1L–39L, in which 1L–9L is the trickling section). The results show the following (1) With the rise of trickling water temperature, the total enthalpy difference of dry air in the roadway increases within a small range and that of humid air goes up nonlinearly. Besides, the increase of trickling water flow rate has an insignificant effect on the sensible heat of the airflow, while it plays a notable role in increasing the latent heat of the airflow. (2) High trickling water temperature results in a higher growth rate of humidity ratio at 19L than those at 29L and 39L in the early stage of thermal water trickling. Meanwhile, sensible heat exchange, which becomes strong after thermal water trickles for over 30 min, complicates the enthalpy difference variation rates of wet air at the three measuring points. (3) The three measuring points in the 19L–39L section all display a process of enthalpy growth with time. In the case of point 39L, the enthalpy difference of humid air surges sharply when the trickling water temperature is 80 ° C or the flow rate is 200 ml/min. The research results boast some reference value for thermal water management and microclimate change forecasting after the airflow passes through a trickling roadway.

This paper proposes a novel U-Net variant using stacked dilated convolutions for medical image segmentation (SDU-Net). SDU-Net adopts the architecture of vanilla U-Net with modifications in the encoder and decoder operations (an operation indicates all the processing for feature maps of the same resolution). Unlike vanilla U-Net which incorporates two standard convolutions in each encoder/decoder operation, SDU-Net uses one standard convolution followed by multiple dilated convolutions and concatenates all dilated convolution outputs as input to the next operation. Experiments showed that SDU-Net outperformed vanilla U-Net, attention U-Net (AttU-Net), and recurrent residual U-Net (R2U-Net) in all four tested segmentation tasks while using parameters around 40% of vanilla U-Net's, 17% of AttU-Net's, and 15% of R2U-Net's., 8 pages MICCAI

Modern deep learning algorithms geared towards clinical adaption rely on a significant amount of high fidelity labeled data. Low-resource settings pose challenges like acquiring high fidelity data and becomes the bottleneck for developing artificial intelligence applications. Ultrasound images, stored in Digital Imaging and Communication in Medicine (DICOM) format, have additional metadata data corresponding to ultrasound image parameters and medical exams. In this work, we leverage DICOM metadata from ultrasound images to help learn representations of the ultrasound image. We demonstrate that the proposed method outperforms the non-metadata based approaches across different downstream tasks., Accept as a workshop paper at AI4AH, ICLR 2020

This Paper conducts effective mining to wind vibration disaster data of power transmission line conductor by using the clustering analysis method. The main influential factors which cause different types of wind vibration disasters are analyzed. This paper laying a foundation for the subsequent correlation analysis and machine learning model. On the basis of K-Means method, the significant degrees of different influential factors on forecast of galloping, wind deviation and sub-span oscillation disaster are studied. The results show that: the span, conductor icing thickness and wind direction are main influential factors of conductor galloping; the included angle between wind and conductor and span are main influential factors of insulator wind vibration; the span, spacer span, and number of spacers are main influential factors of subspan oscillation.

With the development of network communication, time synchronization plays an important role in various network application scenarios. However, different network scenarios exist different characteristics and complexity. Therefore, in order to meet the needs of different application scenarios, various time synchronization schemes are proposed. The Internet is the most common application scenario, and wireless sensor networks(WSNs) are applied under condition-limited scenarios with less nodes. This paper analyzes the existing time synchronization protocols under these two kinds of network from the aspects of accuracy, resource consumption and so on. NTP protocol and PTP protocol are introduced in detail for their outstanding advantages. Time synchronization protocols in WSNs have their own advantages in different aspects, and distributed synchronization mechanisms are mostly better than centralized synchronization mechanisms. We also explore accuracy and security due to importance of them in time synchronization.

The overheating problems of terminal connectors severely threaten the operation of ultra-high voltage projects and cause enormous losses of economy, so a higher currentcarrying reliability of the connectors has an important engineering significance with the development of transmission capacity. In this paper, a bivariate mathematical model of contact resistance as functions of temperature and tightening torque was deduced based on the large current temperature rise tests. To perform such analysis, four typical terminal connectors, namely: overlapping terminal of aluminum plates, overlapping terminal of copper plates, overlapping terminal of plates with clad layer and overlapping terminal of copper rod with aluminum clamp, were chosen for the experiments. The changing rules of DC resistance with different tightening torques and different currents were studied. Then the empirical formula of contact resistance was deduced. Finally, temperature calculations of different terminal connectors were realized to verify the effectiveness of the bivariate mathematical model.

To calculate the stress status of equipment terminal exactly, this paper establishes the overall simulation models of two types of down lead structure systems, and compares them with the test result. The result shows that for cross line - down lead model and truss - down lead model, a relatively good effect has been obtained for both of them by using cable element to simulate conductor, and the error between the terminal stress simulation result and the test result is controlled within 15%, which means that the finite element simulation model of this paper is exact and reasonable.

In this paper, characteristic identification and mechanism analysis on wind vibration disaster of power transmission line are carried out by using the data driving method. To provide technical support for prevention and control of wind vibration disaster to power transmission line, statistics are made in this from three dimensions i.e meteorology & geology, line structure and disaster characteristics to the wind vibration data of power grid conductors. Then frequency distribution characteristics of wind vibration disasters for conductors under different parameter conditions has been analyzed, and the correlation between various factors and wind vibration disasters according to analysis results has also been discussed.

Based on the numerical simulation model of icing, the impact on icing of the current was further investigated. The critical current value with the meteorological conditions of foreign objects, as well as the physical parameters of the aerial cable itself change the law. The calculation results showed that the critical current value depends on the external weather conditions and the aerial cable itself, the physical parameters, and the critical current and the ambient temperature, wind speed, relatively good linear relationship.

Large scale MIMO (multiple-input multiple-output) requires a large number (tens or hundreds) of base station antennas serving a much smaller number of terminals, with large gains in energy efficiency and spectral efficiency compared with traditional MIMO technology. Large scale antennas mean large scale radio frequency (RF) chains. Considering the plenty of power consumption and high cost of RF chains, antenna selection is necessary for large scale MIMO wireless communication systems in both transmitting end and receiving end. In this paper, an energy efficient antenna selection algorithm based on convex optimization is proposed for large scale MIMO wireless communication systems. On the condition that the channel capacity of the cell is larger than a certain threshold, the number of transmit antenna, the subset of transmit antenna and servable mobile terminals (MTs) are jointly optimized to maximize energy efficiency. The joint optimization problem is proved in detail. The proposed algorithm is verified by both mathematical analysis and numerical simulations. According to the simulation results, huge performance gain of energy efficiency is obtained comparing with no antenna selection.

早大学位記番号:新6727

A robust passive UHF RFID backscatter system enhanced with spread spectrum technique is presented in this paper. Due to the weak signal energy of the backscatter chain, traditional RFID backscatter communication is easily affected by the noises, interferences, and interceptions from the environment. To solve the problem, spread spectrum technique is introduced into the backscatter link of RFID system. Simulated results show that this approach largely reduces the Bit Error Rate and improves the system's reliability and security. For hardware realization, spectrum spreading operation is implemented in a RFID tag baseband processor, which is finally applied into a complete RFID tag and fabricated using 0.18um 1P6M CMOS technology. Furthermore, de-spreading operation including the PN acquisition is integrated into a FPGA implementation of RFID reader. Thus, a complete robust RFID backscatter system enhanced with spread spectrum technique is constructed.

Aiming at the expansive application of RFID technology, design considerations are expatiated to a highly-integrated, multimode, power-controllable transmitter in mobile UHF RFID reader with CMOS process. The transmitter consists of digital section, a baseband circuit, an up-conversion mixer and a gain-variable power amplifier. Multiple data modes can be generated. The direct-conversion RF section is proposed to minimize the off-chip components and provide a low-cost, highly efficient solution. The RF power amplifier can achieve a maximum output power of 26 dBm and maintain a power-added efficiency (PAE) higher than 40% over the 18 dBm-26 dBm output power range. Feasibility of the transmitter is validated by ADS simulator.

A low cost fully integrated single-chip UHF radio frequency identification (RFID) reader SoC for short distance handheld applications is presented. The SoC integrates all building blocks—including an RF transceiver, a PLL frequency synthesizer, a digital baseband and an MCU—in a 0.18 μm CMOS process. A high-linearity RX front-end is designed to handle the large self-interferer. A class-E power amplifier with high power efficiency is also integrated to fulfill the function of a UHF passive RFID reader. The measured maximum output power of the transmitter is 20.28 dBm and the measured receiver sensitivity is −60 dBm. The digital baseband including MCU core consumes 3.91 mW with a clock of 10 MHz and the analog part including power amplifier consumes 368.4 mW. The chip has a die area of 5.1 × 3.8 mm2 including pads.

A low cost integrated transceiver for mobile UHF passive RFID reader applications is implemented in a 0.18- m CMOS process. The transceiver contains an OOK modulator and a power amplifier in the transmitter chain, an IQ direct-down converter, variable-gain amplifiers, channel-select filters and a 10-bit ADC in the receiver chain. The measured output P1dB power of the transmitter is 17.6 dBm and the measured receiver sensitivity is -70 dBm. The on-chip integer N synthesizer achieves a frequency resolution of 200 kHz with a phase noise of -104 dBc/Hz at 100 kHz frequency offset and -120.83 dBc/Hz at 1 MHz frequency offset. The transmitter, the receiver and the frequency synthesizer consume 201.34, 25.3 and 54 mW, respectively. The chip has a die area of 4 2.5mm 2