Search

Neural network search (NAS) projects a promising direction to automate the design process of efficient and powerful neural network architectures. Nevertheless, the NAS techniques have to dynamically generate a large number of candidate neural networks, and iteratively train and evaluate these on-line generated network architectures, thus they are extremely time-consuming even when deployed on large GPU clusters, which dramatically hinders the adoption of NAS. Though recently there are many specialized architectures proposed to accelerate the training or inference of neural networks, we observe that existing neural network accelerators are typically targeted at static neural network architectures, and they are not suitable to accelerate the evaluation of the dynamical neural network candidates evolving during the NAS process, which cannot be deployed onto current accelerators via the off-line compilation. To enable rapid and energy-efficient NAS in compact singlechip solutions, we propose NASA, a specialized architecture for one-shot based NAS acceleration. It is able to generate, schedule, and evaluate the candidate neural network architectures for the target machine learning workload with high speed, significantly alleviating the processing bottleneck of one-shot NAS. Motivated by the observation that there are considerable computation sharing opportunities among the different neural network candidates generated in one-shot NAS, NASA is equipped with an on-chip network fusion unit to remove the redundant computation during the network mapping stage. In addition, the NASA accelerator can partition and re-schedule the candidate neural network architectures at fine-granularity to maximize the chance of data reuse and improve the utilization of the accelerator arrays integrated to accelerate network evaluation. According to our experiments on multiple one-shot NAS tasks, NASA achieves 33.52X performance speedup and 214.33X energy consumption reduction on average when compared to a CPU-GPU system.

Summary: Background: Improving the quality of hospital antibiotic use is a major goal of WHO's global action plan to combat antimicrobial resistance. The WHO Essential Medicines List Access, Watch, and Reserve (AWaRe) classification could facilitate simple stewardship interventions that are widely applicable globally. We aimed to present data on patterns of paediatric AWaRe antibiotic use that could be used for local and national stewardship interventions. Methods: 1-day point prevalence survey antibiotic prescription data were combined from two independent global networks: the Global Antimicrobial Resistance, Prescribing, and Efficacy in Neonates and Children and the Global Point Prevalence Survey on Antimicrobial Consumption and Resistance networks. We included hospital inpatients aged younger than 19 years receiving at least one antibiotic on the day of the survey. The WHO AWaRe classification was used to describe overall antibiotic use as assessed by the variation between use of Access, Watch, and Reserve antibiotics, for neonates and children and for the commonest clinical indications. Findings: Of the 23 572 patients included from 56 countries, 18 305 were children (77·7%) and 5267 were neonates (22·3%). Access antibiotic use in children ranged from 7·8% (China) to 61·2% (Slovenia) of all antibiotic prescriptions. The use of Watch antibiotics in children was highest in Iran (77·3%) and lowest in Finland (23·0%). In neonates, Access antibiotic use was highest in Singapore (100·0%) and lowest in China (24·2%). Reserve antibiotic use was low in all countries. Major differences in clinical syndrome-specific patterns of AWaRe antibiotic use in lower respiratory tract infection and neonatal sepsis were observed between WHO regions and countries. Interpretation: There is substantial global variation in the proportion of AWaRe antibiotics used in hospitalised neonates and children. The AWaRe classification could potentially be used as a simple traffic light metric of appropriate antibiotic use. Future efforts should focus on developing and evaluating paediatric antibiotic stewardship programmes on the basis of the AWaRe index. Funding: GARPEC was funded by the PENTA Foundation. GARPEC-China data collection was funded by the Sanming Project of Medicine in Shenzhen (SZSM2015120330). bioMérieux provided unrestricted funding support for the Global-PPS.

This paper presents the design and characteristics of a new two-dimensional nonresonant tertiary motion generator which is based on the flextensional structure. A tool holder connects two perpendicularly placed flextensional actuators with flexure hinges which decouple the motion outputs from the two actuators. Piezoelectric stacks, which are preloaded through precision screws, are used to provide input displacements. By balancing the requirements of driving current, stiffness, and the displacement amplification ratio, the proposed design is targeted to operate at above 10 kHz with two-dimensional vibrations amplitude of 10 μm in each direction. Technical difficulties in driving a nonresonant mode piezoelectric actuator at a high frequency are discussed. The solutions and optimization procedures are presented in this paper. The design is optimized by finite-element simulation; and the results are presented and verified by our prototype design.

The impinging streams mixer is a new type micromixer. The cavitation phenomenon occurring in the mixers with T-shaped impinging streams (TS), conical impinging streams (CIS), vortex streams(VS) were investigated, respectively. The distribution of flow field in the mixer was simulated and calculated by commercial software Fluent 6.2.1. The results showed that under the same working conditions, a more obvious hydrodynamic cavitation may occur in the CIS type impinging stream than that in the CIS type or the VS type, and the vortex flow lead to an extension of the material residence time in the mixer. The distribution of turbulent kinetic energy and gas holdup were obtained by numerical simulating hydrodynamic cavitation under different entrance pressure conditions. It is showed that when the outlet pressure keeps a constant value, hydrodynamic cavitation can be enhanced by increasing the entrance pressure. The above research can be contributed to the producing of biodiesel and the solving of the key technical problem of oil and alcohol heterogeneous mixing.

By using online measurement and inspection, the efficiency of production in product manufacturing can be greatly improved. However, most of the available methods for the measurement and inspection of parts are offline in practice, which results in either time consuming or expensive inefficiencies. In this paper, a vision-based inspection approach is proposed, which is designed to measure and inspect parts with 2.5 dimensional shapes online within an appropriate accuracy. The hardware of the system includes: a scanner that is utilized to capture an image of the part to be inspected, a control device that manages the scanner so that it can capture the image at the appropriate time, a backlighting table that provides enough light to help acquire a clearer image, and a box to prevent outside light interference. The software of the system, implemented by using a Matlab toolkit, was used to perform the data processes, which consist of several modules including, but not limited to image segmentation, geometric feature extraction and calibration. To verify the validity of the measurement system, one simple part was used for testing purposes. The technique's efficiency was demonstrated by the comparison of the measured parameters with their actual values.

According to the basic principle of integrated navigation, the algorithm of Unscented Kalman Filter is used in current study. Both the simulation of Beidou/SINS Integrated Navigation system and a car running experiment are proposed in this paper. For the car experiment, using an autopilot for Guided Bomb which equipped with a receiver of Beidou and a Micro Inertial Measure Unit (MIMU), and then the experiment is proceeded on a car. The result of this experiment shows that using the algorithm of UKF fusion of the output information of Beidou Navigation System and SINS can effectively estimate the state errors. Under the process of feedback compensation, it can enormously improve the precision and the data refresh rate of the integrated navigation information. Under the condition of the hardware and software mentioned in this paper, the accuracy of location can reach 6 meters, and the data refresh rate can be improved to be 100Hz comparing with 1Hz that the Beidou gives. The accuracy indexes showed in this experiment can basically satisfy the requirement of a Precision Guided Bomb.

In this paper, a novel sliding mode method is proposed to suppress the chattering phenomenon with unstable surfaces. Unlike the conventional sliding mode control methods, these unstable surfaces are utilized to design an available sliding mode structure. The main idea of this method is to design a sliding mode controller following a new reaching condition with unstable auxiliary surfaces. Based on this idea, we might find a new way to design a large number of different sliding mode controllers to solve the chattering problem of sliding mode method. The simulation results are given to illustrate the benefits and properties of the proposed method.

Aging characteristics and mechanical properties of commercial 7xxx series Al composites were investigated from viewpoint of ceramic contents. After sintering process, sintered densities of blended and composite powder were 95 and 97%, respectively. Each part was solution-treated at for 60 min and aged . And two-step aging was also performed form . The aging behavior of the sintered composite pow-der was different from that of sintered blended powder. The peak aging time of the composite was rapid as well due to strain. Before aging, mechanical properties of sintered composite powder was significantly higher than that of sintered blended powder. These increments of properties were directly affected by ceramic particles. However, after aging, incremental rate of mechanical properties was slowed in the composite.

7xxx series Al alloy has the most attractive properties including its excellent high specific strength, stress corrosion cracking and corrosion-resistance. However, in case of the Al-Zn system, the liquid phase has a transient aspect because of the high solid solubility of Zn in Al. Therefore, transient liquid phase sintering behavior was observed during the sintering process. And the amount of liquid and its duration were influenced by the process variables including heating rate and final sintering temperature. At high heating rates(), the liquid fraction increased during sintering because diffusion was minimized and therefore local saturation could easily occur. The sintered density increased with increasing heating rate.

The bi-materials composed of and its composite reinforced with SiC particles were prepared by ball-milling and subsequent sintering process. The size of powder in Al-Mg/SiCp mixture decreased with increasing ball-milling time, it was saturated above 30 h when the ball and powder was in the ratio of 30 to 1. Both powders mixture and mixture were compacted under a pressure of 350MPa and were bonded by sintering at temperatures ranging from 873K to 1173K for 1-5h. At 873k, the sound bi-mate-rials could not be obtained. In contrast, the bi-materials with the macroscopically well-bonded interface were obtained at higher temperatures than 873K. The length of well-bonded interface became longer with increasing temperature and time, indicating the improved contact in the interface between unreinforced Al-Mg part and Al-Mg/SiCp composite part. The relative density in the bi-materials increased as the sintering temperature and time increased, and the bi-materials sintered at 1173K for 5h showed the highest density.

Peer instruction (PI) is an effective interactive approach to teaching and learning that has principally been used to modify the experience of learning in traditional physics lecture settings. This article further illustrates how the concept of PI can be effectively applied in the physics student laboratory setting. The setting used is a laboratory task that calls for the measurement of the effective mass of the spring of a Jolly balance. Through PI the students gain a better understanding of what is meant by the construct 'effective mass of a spring', and thereby competently work out how the mass, shape, wire diameter, and number of turns of the spring can all affect the effective mass of the spring. Furthermore, using stopwatches the students were also able to appreciate how recorded times at the equilibrium position had greater uncertainty than measurements made at the maximum displacement. This led to their calculations of the effective mass of the spring being impressively close to the theoretical value. Such laboratory tasks are extremely challenging to introductory level students and the success attained by the students in this study indicates that there is much potential in the application of PI in laboratory settings. PI should be used to teach in the laboratory and results should be reported in order for our community to build on these experiences. This article is a contribution to that effort.

Glycerin has been used as a drug carrier/depot, but never with local anaesthestics. This study was an attempt to use the slow drug release mechanism to prolong the anaesthetic effects of bupivacaine in epidural block. Twenty-seven adults with cancer pain were prospectively selected according to their primary lesions and problems, but their allocation to study groups was randomized. Group I (n = 13), received 5 ml bupivacaine, 0.125% in normal saline via a previous implanted epidural catheter. When the pain returned to its original intensity, the same amount of the same strength anaesthetic dissolved in 50% glycerin was given via the same catheter. Group II (n = 14) received the same solutions, but in the reverse order. Also five patients in each group received plain 50% glycerin prior to administration of the anaesthetic solutions to serve as controls. The pharmacological effects were assessed by the blinded observers. Analgesia produced with glycerin solution was prolonged compared with the saline solution (12.2 vs 7.2 and 11.6 vs 7.4 hr, P < 0.01). The order of giving the solution did not produce any differences. Plan 50% glycerin did not produce analgesic effects. Neither motor blockade nor other adverse effects or complications were observed in either group. It was concluded that 0.125% bupivacaine in 50% glycerin administered epidurally is not neurotoxic. The prolongation of analgesia observed is attributed to the slow release of bupivacaine from the glycerin base which functions as drug depot. In addition to relief of chronic pain, this novel approach may have other clinical applications such as the relief of labour or postoperative pain.

This paper presents a optimal control and allocation strategy for Internal combustion-Linear Generator integrated Power System (ICLGIPS), the main objective of the strategy is to improve efficiency of the system. Linear generator (LEG) works alternately as electromotor and generator during one work period, energy storage device converts and transfers energy between power source and LEG, energy conversion efficiency and transmission efficiency is vital in the device. By analyzing the system working characteristics, transfer module and conversion module of energy storage device are designed, this design program effectively reduces size and cost of energy storage device and makes the system work with greater reliability, so efficiency of the system is improved. The energy flow control scheme is designed by studying the characteristics of energy flow. Through analysis and calculation, the entire energy storage device model is build, the model is simulated using MATLAB/SIMULINK. The simulation results show that the optimal control and allocation strategy more effectively improve efficiency of energy flow, reaching more than 95%, meet the system requirements of environmental protection and energy saving, and have good theoretical value.

This paper was presented at the 2nd Micro and Nano Flows Conference (MNF2009), which was held at Brunel University, West London, UK. The conference was organised by Brunel University and supported by the Institution of Mechanical Engineers, IPEM, the Italian Union of Thermofluid dynamics, the Process Intensification Network, HEXAG - the Heat Exchange Action Group and the Institute of Mathematics and its Applications. Angiogenesis, formation of new blood vessels, is a complex but critical phenomenon. In particular, it is regulated by different angiogenic factors. Nitric oxide (NO) is also a very well known biological mediator involved in vascular physiology. This study focuses on relationships between the effect of angiogenin, a major angiogenic factor, and extracellular NO release. NO concentration was sensed electrochemically using a fibronectin coated multiple microelectrode array. Angiogenin was shown to increase NO levels, thus triggering nitric oxide synthase (NOS) activity. Angiogenin reactive pathway being very complex, we have used various selective inhibitors of angiogenin to investigate the mechanism leading to NO production. Neomycin, an antibiotic blocking nuclear translocation, inhibited angiogenin effect on NOS. This result demonstrates that angiogenin activates NOS by interacting with the cell nucleus. This study is funded by Medermica Ltd; the DIUS; KICOS (K20602000681-08B0100-02210); the Korea Science and Engineering Foundation (M10749000231-08N4900-23110); and the Korea Biotech R&D Group of Next-Generation Growth Engine Project (F104AB010004-08A0201-00410).

A method which improve the quality of the reconstructed image obtained from the wavelet transform which is appropriate for the human visual system using non-linear quincunx interpolation filtering is proposed. A straightforward method to extend the 1-D wavelet basis to the 2-D wavelet basis is the separable wavelet transform. Approximation images and detail images are obtained with a multiresolution image representation. Each detail image has horizontal direction edge, vertical direction edge and diagonal direction edge information respectively. Non-linear interpolation filtering of each detail image improve the quality (PSNR) of the reconstructed image by some 1-2 dB.

In this paper, we present a simple parallel architecture for Discrete Wavelet Transform (DWT). Efficient computation of the pyramid algorithm for the computing of the discrete wavelet transform is possible due to the similarity between computation results of each octave. By using similarity, we separated the filter into 2 parts, an even filter and an odd filter. 1 octave and other octave computation are performed in the even and odd filters at the same time. The proposed architecture has following features. (1) Critical path is 1 multiplier and 1 adder; (2) the number of required registers is 1+J*([L/sub h//2]-1)+1+J*([L/sub 1//2]-1)+J, where J is the number of octaves, L/sub h/ is length of the highpass filter and L/sub 1/ is length of the lowpass filter.