Search

We explore conditions for when the gradient of a deep declarative node can be approximated by ignoring constraint terms and still result in a descent direction for the global loss function. This has important practical application when training deep learning models since the approximation is often computationally much more efficient than the true gradient calculation. We provide theoretical analysis for problems with linear equality constraints and normalization constraints, and show examples where the approximation works well in practice as well as some cautionary tales for when it fails., 10 pages, 4 figures, ICML 2023 workshop on Differentiable Almost Everything

Multimodal alignment facilitates the retrieval of instances from one modality when queried using another. In this paper, we consider a novel setting where such an alignment is between (i) instruction steps that are depicted as assembly diagrams (commonly seen in Ikea assembly manuals) and (ii) video segments from in-the-wild videos; these videos comprising an enactment of the assembly actions in the real world. To learn this alignment, we introduce a novel supervised contrastive learning method that learns to align videos with the subtle details in the assembly diagrams, guided by a set of novel losses. To study this problem and demonstrate the effectiveness of our method, we introduce a novel dataset: IAW for Ikea assembly in the wild consisting of 183 hours of videos from diverse furniture assembly collections and nearly 8,300 illustrations from their associated instruction manuals and annotated for their ground truth alignments. We define two tasks on this dataset: First, nearest neighbor retrieval between video segments and illustrations, and, second, alignment of instruction steps and the segments for each video. Extensive experiments on IAW demonstrate superior performances of our approach against alternatives., Comment: Project website: https://academic.davidz.cn/en/publication/zhang-cvpr-2023/

The long-tailed recognition (LTR) is the task of learning high-performance classifiers given extremely imbalanced training samples between categories. Most of the existing works address the problem by either enhancing the features of tail classes or re-balancing the classifiers to reduce the inductive bias. In this paper, we try to look into the root cause of the LTR task, i.e., training samples for each class are greatly imbalanced, and propose a straightforward solution. We split the categories into three groups, i.e., many, medium and few, according to the number of training images. The three groups of categories are separately predicted to reduce the difficulty for classification. This idea naturally arises a new problem of how to assign a given sample to the right class groups? We introduce a mutual exclusive modulator which can estimate the probability of an image belonging to each group. Particularly, the modulator consists of a light-weight module and learned with a mutual exclusive objective. Hence, the output probabilities of the modulator encode the data volume clues of the training dataset. They are further utilized as prior information to guide the prediction of the classifier. We conduct extensive experiments on multiple datasets, e.g., ImageNet-LT, Place-LT and iNaturalist 2018 to evaluate the proposed approach. Our method achieves competitive performance compared to the state-of-the-art benchmarks.

We explore two differentiable deep declarative layers, namely least squares on sphere (LESS) and implicit eigen decomposition (IED), for learning the principal matrix features (PMaF). It can be used to represent data features with a low-dimensional vector containing dominant information from a high-dimensional matrix. We first solve the problems with iterative optimization in the forward pass and then backpropagate the solution for implicit gradients under a bi-level optimization framework. Particularly, adaptive descent steps with the backtracking line search method and descent decay in the tangent space are studied to improve the forward pass efficiency of LESS. Meanwhile, exploited data structures are used to greatly reduce the computational complexity in the backward pass of LESS and IED. Empirically, we demonstrate the superiority of our layers over the off-the-shelf baselines by comparing the solution optimality and computational requirements., Comment: 16 pages, 7 figures, 10 tables, accepted to the differentiable almost everything workshop, ICML, 2023

The generation of three-dimensional (3D) medical images can have great application potential since it takes into account the 3D anatomical structure. There are two problems, however, that prevent effective training of a 3D medical generative model: (1) 3D medical images are very expensive to acquire and annotate, resulting in an insufficient number of training images, (2) a large number of parameters are involved in 3D convolution. To address both problems, we propose a novel GAN model called 3D Split&Shuffle-GAN. In order to address the 3D data scarcity issue, we first pre-train a two-dimensional (2D) GAN model using abundant image slices and inflate the 2D convolution weights to improve initialization of the 3D GAN. Novel 3D network architectures are proposed for both the generator and discriminator of the GAN model to significantly reduce the number of parameters while maintaining the quality of image generation. A number of weight inflation strategies and parameter-efficient 3D architectures are investigated. Experiments on both heart (Stanford AIMI Coronary Calcium) and brain (Alzheimer's Disease Neuroimaging Initiative) datasets demonstrate that the proposed approach leads to improved 3D images generation quality with significantly fewer parameters., 10 pages

We present NeRFEditor, an efficient learning framework for 3D scene editing, which takes a video captured over 360{\deg} as input and outputs a high-quality, identity-preserving stylized 3D scene. Our method supports diverse types of editing such as guided by reference images, text prompts, and user interactions. We achieve this by encouraging a pre-trained StyleGAN model and a NeRF model to learn from each other mutually. Specifically, we use a NeRF model to generate numerous image-angle pairs to train an adjustor, which can adjust the StyleGAN latent code to generate high-fidelity stylized images for any given angle. To extrapolate editing to GAN out-of-domain views, we devise another module that is trained in a self-supervised learning manner. This module maps novel-view images to the hidden space of StyleGAN that allows StyleGAN to generate stylized images on novel views. These two modules together produce guided images in 360{\deg}views to finetune a NeRF to make stylization effects, where a stable fine-tuning strategy is proposed to achieve this. Experiments show that NeRFEditor outperforms prior work on benchmark and real-world scenes with better editability, fidelity, and identity preservation., Comment: Project page: https://chuny1.github.io/NeRFEditor/nerfeditor.html

Photoacoustic Doppler flow measurement based on continuous wave laser excitation owns the merit of clearly presenting the Doppler power spectra. Extending this technique to dual wavelengths can gain the spectral information of the flow sample extra to the flow speed information. An experimental system with two laser diodes respectively operated at 405 nm and 660 nm wavelengths is built and the flow measurement with black and red dyed polystyrene beads is performed. The measured Doppler power spectra can vividly reflect the flow speed, the flow direction, as well as the bead color. Since it is straightforward to further apply the same principle to multiple wavelengths, we can expect this type of spectroscopic photoacoustic Doppler flow measurement will be developed in the near future which will be very useful for studying the metabolism of the slowly moving red blood cell inside microvessels.

Regarding a load-sensitive system with after-valve compensation, in the case of a single action or composite action, theoretically, the flow of each channel neither varies with the load pressure of this channel, nor is it affected by the flow of other channels, and the anti-interference performance is good. In fact, due to the influence of factors such as the flow passage of the valve body, the matching between the pressure compensation valve and the main spool, its flow control and anti-interference performance are difficult to achieve good results, which has a great impact on the synchronization, rapid action impact and micro motion characteristics of the main engine of construction machinery. This paper studies the simplification and analysis of the system pressure drop diagram, the matching characteristics between the pressure compensation valve and the main valve, accurate modeling and simulation technology, solves the problems of flow fine control and anti-interference performance, improves the flow control accuracy of a single action and compound action, and enhances the anti-interference performance. The real vehicle test shows that the self-developed multi-way valve for compensating for load sensitivity behind the valve is installed on the 6T small excavation. The initial position of the excavator’s track is aligned along a straight line. After the test runs for 20 meters, the track offset is measured. The fast forward offset is only 20mm, and the slow forward offset is only 7mm.

Generative models such as generative adversarial networks and autoencoders have gained a great deal of attention in the medical field due to their excellent data generation capability. This paper provides a comprehensive survey of generative models for three-dimensional (3D) volumes, focusing on the brain and heart. A new and elaborate taxonomy of unconditional and conditional generative models is proposed to cover diverse medical tasks for the brain and heart: unconditional synthesis, classification, conditional synthesis, segmentation, denoising, detection, and registration. We provide relevant background, examine each task and also suggest potential future directions. A list of the latest publications will be updated on Github to keep up with the rapid influx of papers at https://github.com/csyanbin/3D-Medical-Generative-Survey., Comment: A survey on the 3D brain and heart volume generative models

Vibration signal, as an important means for diesel engine condition detection and fault diagnosis, has attracted attention for many years. In traditional vibration signal analysis, most processing methods are for single-channel data. However, single-channel vibration signal cannot reflect the operating information of the diesel engine comprehensively because diesel engine vibration is coupled by multiple source signals. This paper proposes the MVMD band energy method for fault diagnosis by four channels of vibration signals. First, the original multivariate signals are decomposed adaptively by MVMD, which obtains a series of components with modal alignment. Then, the band energy values of each measuring point are calculated as the fault characteristics. Finally, SVM is used to realize the diagnosis and identification of diesel engine misfire. The working conditions have a great influence on the vibration signal of the cylinder. In order to obtain the best diagnostic working conditions, six working conditions are set for testing. The result shows that the fault identification rate is highest under the 1500 rpm and 50% load working condition. The fault recognition rate of this method reaches more than 99%, which is superior to the other four common methods.

In order to improve the reliability and service life of vehicle and diesel engine, the fatigue life prediction of the piston in a heavy diesel engine was studied by finite element analysis of piston, experiment data of aluminum alloy, fatigue life model based on energy dissipation criteria, and machine learning algorithm. First, the finite element method was used to calculate and analyze the temperature field, thermal stress field, and thermal–mechanical coupling stress field of the piston, and determine the area of heavy thermal and mechanical load that will affect the fatigue life of the piston. Second, based on the results of finite element calculation, the creep–fatigue experiment of 2A80 aluminum alloy was carried out, and the cyclic response characteristics of the material under different loading conditions were obtained. Third, the fatigue life prediction models based on energy dissipation criterion and twin support vector regression are proposed. Then, the accuracy of the two models was verified using experiment data. The results show that the model based on the twin support vector regression is more accurate for predicting the material properties of aluminum alloy. Based on the established life prediction model, the fatigue life of pistons under actual service conditions is predicted. The calculation results show that the minimum fatigue life of the piston under plain condition is 2113.60 h, and the fatigue life under 5000 m altitude condition is 1425.70 h.

The pipeline is the main channel for concrete pumping in construction, and lightweight, safe, long service life pipeline is very important for concrete pump trucks. This paper presents the design and experimental study on hybrid structure of carbon fiber reinforced polymer (CFRP) filament-winded strengthened circular long steel pipe. The filament-winded CFRP contains three layers: the inner layer for stress sharing of brittle steel tube, the middle layer for connecting to ductile steel flanges and the outer layer for explosion-proof and protection. With calculation of the hybrid structure stress, the design guideline of inner and outer layer CFRP is presented. Based on the design and test of the connection structure between pipe flange and CFRP, this paper conducted an optimization on the filament winding patterns of the middle layer. The experiment result shows the new hybrid structure pipeline can bear the pressure of 17 MPa and the axial loading of 210 kN, and the working requirements of concrete pumping can be satisfied. The test result indicates that the hybrid structure pipeline reduces 43% mass compared with the traditional double-layer steel pipe, and also the service life is 50% increased so as to the explosion-proof safety performance of concrete pump trucks is obviously improved.

Fatigue properties of Q500D advanced high strength steel (AHSS) with a medium plate thickness welded by both hybrid plasma arc welding (HPAW) process and regular GMAW process were investigated. Compare to regular GMAW process, HPAW joint showed longer fatigue life. Both optical microscope and electron microscope analysis were carried out to investigate the fracture origin. It showed that the HPAW process provides less stress concentration and less facture origin than GMAW, which guaranteed the fatigue life of weld joint.

Purpose Based on the significance of ectomycorrhizae (ECM) and increased publication activity on this subject, it was decided to carry out a bibliometric analysis of scientific outputs in this area. The purpose of this study is to reveal the research trends of scientific outputs on ECM for the past 30 years and provide a potential guide for future research. Design/methodology/approach A method of bibliometric analysis was performed, based on the online version of the Science Citation Index Expanded, Web of Science, from 1986 to 2017. The authors evaluated the publication types, languages, source countries, journals, the patterns of publication outputs, most-cited articles, most-productive authors, institutional distributions, subject categories, high-frequency keywords and keywords plus and high-frequency terms in the title and abstract of ectomycorrhizal research. Keywords, keywords plus and high-frequency terms in the title and abstract were analyzed via VOSviewer to illustrate the extent of co-occurrence. This study further describes the recent research priority or hotspots and reveals the research trends. Findings From 1986 to 2017, the publication output on ECM showed a rising trend; the number of articles has rapidly increased after 2003. Based on co-occurrence analysis for keywords, keywords plus and terms in the title and abstract, “ectomycorrhizal fungi” is the most popular keyword and keywords plus; “concentration” is the most high-frequency terms in the title and abstracts. Plant biology, mycology and ecology are the hotspots in the ectomycorrhizal research. Ectomycorrhizal taxonomy, the molecular mechanisms of ectomycorrhizal symbioses and the common mycorrhizal networks are the future direction. Originality/value A bibliometric analysis has been carried out to analyze the trends of ECM research with 30 years. This study provides a potential guide for future research related to ectomycorrhizae.

In order to study the driving state of heavy vehicles, two approaches are employed hereby to establish digital twin models for analyzing the applicable scopes of the models and conducting a predictive study. To begin with, the operating parameters and the state of the vehicles are measured using instruments and apparatuses. Then, relying on a GP and deep convolutional neural network (DCNN), two digital twin models of vehicles driving state are established, respectively, which set the transmission system and power system parameters as well as weather conditions as input parameters; and vehicle running speed and torque value as output ones. Both digital twin models consider the physical rule of the vehicle to avoid overfitting in the training. The analytical results indicate that the GP-based digital twin model appears more accurate in predicting the driving parameters of the vehicles, whereas the model based on the DCNN has better convergence precision within a short span of time. The vehicle-specific digital twin models set up in this paper lay a foundation for subsequent optimization of vehicle driving state and realization of digital twin-physical entity interaction.

In order to increase the reliability and service life of piston in a heavy-duty diesel engine, the geometric structure of piston was optimized based on its maximum temperature and maximum coupling stress. To begin with, the boundary conditions of thermal and stress fields are calculated, which include the heat produced by the combustion in cylinder, the friction-induced heat, and the heat transferred to cooling system. Then, the finite element model was established to calculate and analyse the temperature and thermal-mechanical coupling stress fields of the piston. By combining this simulation model with orthogonal experimental design methods, computations and analyses were performed to determine how the five geometric parameters (depth of intake and exhaust valve grooves, radius of valve grooves transition, radius of top of valve grooves, height of first piston ring groove, and depth of piston ring groove) influence the two evaluation indicators (maximum temperature and maximum stress of piston). Subsequently, using the proposed ABC-OED- FE (artificial bee colony, orthogonal experiment design, and fitting equations) method, the fitting equations between the geometric parameters and evaluation indicators were determined. Taking the minimum values of two evaluation indicators of piston as optimization objectives, artificial bee colony method was run to determine the values of parameters. At last, the two evaluation indicators of the optimized piston were computed. The results indicate that, after optimization, the maximum temperature of piston decreases to be 16.05 K and the maximum stress decreases to be 13.54 MPa. Both temperature and stress conditions of the optimized piston had been improved, which demonstrates the effectiveness of the optimization and the validity of the algorithm.

University of Technology Sydney. Faculty of Engineering and Information Technology. The recent success of convolutional neural networks (CNNs) relies on a large amount of annotated training data. However, many research problems suffer from the scarcity of labeled data, since annotating a large number of data is time-consuming or infeasible. This dissertation focuses on learning with limited labeled data and addresses two problems: few-shot classification and object matching. For few-shot classification, a transductive propagation network (TPN) is first proposed to deal with the low-data issue. The idea is learning to propagate labels from labeled instances to unlabeled ones by exploiting the manifold structure of the data. Then, online feature selection with imbalanced streaming data, as a special few-shot problem, is tackled by the proposed adaptive sparse confidence-weighted (ASCW) algorithm. This algorithm utilizes the confidence-weighted (CW) learning to explore the feature correlation and maintains multiple confidence-weighted learners with different costs to address the imbalanced issue. For object matching, since the labeled matching pairs are usually scarce, finding the potential matching among unpaired objects is important. Based on this idea, two models are proposed to solve object matching with limited labeled data. First, a squared-loss mutual information (SMI) estimator is proposed to utilize a small number of paired samples and the available unpaired ones. The estimator is formulated with optimal transport and quadratic programming in an iterative way. Second, the specific object matching problem, namely semantic correspondence, can be solved in a unified optimal transport framework. The many to one matching and background matching issues are well addressed in the proposed framework. To evaluate the effectiveness of the aforementioned algorithms with limited labeled data, extensive experiments are conducted on various benchmark datasets, ranging from the UCI machine learning repository, few-shot image classification datasets, semantic correspondence datasets, etc.

Previous photoacoustic flow measurements have been demonstrated with various modalities including Doppler frequency shift measurement from intensity modulated continuous laser wave and pulsed photoacoustic Doppler cross- correlation from nanosecond laser pulse. Here, we propose and demonstrate a new type of photoacoustic Doppler velocimetry based on a hybrid approach: we applied a supercontinnum laser source with picosecond pulse and 40 MHz repetition rate, and then use electo-optic modulator to modulate the intensity at 1 MHz. This method can potentially improve the signal to noise ratio due to pulse excitation of photoacoustic signal, and at the same time to achieve the high resolution of Doppler frequency shift due to the intensity modulation. With this method, we measured the moving speeds of various samples at the range of 0.1-1.6mm/s and obtained accurate results. The extra advantage of using supercontinnum laser pulses is that the spectral range can be chosen by using spectral filter thus to realize the multispectral measurement. To demonstrate this, we set the spectral filter working at five ranges at 500nm, 550nm, 600nm, 650nm, 700nm center wavelengths all with 100nm spectral width and measured different materials as moving targets fixed on the translation stage, including such as graphene thin layer, black tape, red and black dyed microspheres.

With the rapid development of communication and transportation technologies, the urban area is increasingly becoming an ever more dynamic, comprehensive, and complex system. Meanwhile, functional polycentricity as a distinctive feature has been characterizing urban areas around the world. However, the spatial structure of the urban area has yet to be fully comprehended from a dynamic perspective, and understanding the spatial organization of polycentric urban regions (PUR) is crucial for issues related to urban planning, traffic control, and urban risk management. The analysis of polycentricity strongly depends on the spatial scale. In order to identify functional polycentricity at the intra-unban scale, this paper presents a traffic flow-embedded and topic modeling-based methodology framework. This framework was evaluated on real-world datasets from the Wujiang district, Suzhou, China, which contains 151,419 records of taxi trajectory data and 86,036 records of points of interest (POI) data. This paper provides a novel approach to examining urban functional polycentricity via combining urban function distribution and spatial interactions. This proposed methodology can help urban authorities better understand urban dynamics in terms of function distribution and internal connectedness and facilitate urban development in terms of urban planning and traffic control.

In order to solve issues concerning performance induction and in-cylinder heat accumulation of a certain heavy-duty diesel engine in a plateau environment, working state parameters and performance indexes of diesel engine are calculated and optimized using the method of artificial neural network and genetic algorithm cycle multi-objective optimization. First, with an established diesel engine simulation model and an orthogonal experimental method, the influence rule of five performance indexes affected by five working state parameters are calculated and analyzed. Results indicate the first four of five working state parameters have a more prominent influence on those five performance indexes. Subsequently, further calculation generates correspondences among four working state parameters and five performance indexes with the method of radial basis function neural network. The predicted value of the trained neural network matches well with the original one. The approach can fulfill serialization of discrete working state parameters and performance indexes to facilitate subsequent analysis and optimization. Next, we came up with a new algorithm named RBFNN &amp, GACMOO, which can calculate the optimal working state parameters and the corresponding performance indexes of the diesel engine working at 3700 m altitude. At last, the bench test of the diesel engine in a plateau environment is employed to verify accuracy of the optimized results and the effectiveness of the algorithm. The research first combined the method of artificial neural network and genetic algorithm to specify the optimal working state parameters of the diesel engine at high altitudes by focusing on engine power, torque and heat dissipation, which is of great significance for improving both performance and working reliability of heavy-duty diesel engine working in plateau environment.

In order to improve reliability and fatigue life of cylinder gaskets in heavy duty diesel engine, several methods and algorithms are applied to optimize operating factors of gaskets. Finite element method is utilized to figure out and analyze the temperature fields, thermal-mechanical coupling stress fields, and deformations of gasket. After determining the maximum values of three state parameters, the orthogonal experimental design method is adopted to analyze the influence rules of five operating factors on three state parameters of the gaskets and four factors which most significantly affect these state parameters are determined. Then, the method which uses operating factors to predict state parameters is established on the application of hybrid neuron network based on partial least squares regression and deep neural network. The comparison results between the predicted values and real values verified the accuracy of the hybrid neuron network method. Based on artificial bee colony algorithm, improvement is attached to the way three kinds of grey wolves locate preys in grey wolf algorithm and the way how using different hierarchy wolfs in grey wolf algorithm to determine three weight coefficients and the location of prey is put forward with. The method using artificial bee colony algorithm to optimize the grey wolf algorithm is called ABC and GWO. The proposed HNN and the ABC and GWO method are applied to work out operating factors values which correspond to optimal state parameters of gasket, and the gaskets are optimized according to the optimal values. It has been demonstrated by finite element analysis results that maximum temperature, maximum coupling stress, and the maximum deformation decrease to 6 K, 12.57 MPa, and 0.0925 mm compared to the original values, respectively, which proves the accuracy of the algorithm and the validity of the improvement.

Optimal transport is a machine learning problem with applications including distribution comparison, feature selection, and generative adversarial networks. In this paper, we propose feature-robust optimal transport (FROT) for high-dimensional data, which solves high-dimensional OT problems using feature selection to avoid the curse of dimensionality. Specifically, we find a transport plan with discriminative features. To this end, we formulate the FROT problem as a min--max optimization problem. We then propose a convex formulation of the FROT problem and solve it using a Frank--Wolfe-based optimization algorithm, whereby the subproblem can be efficiently solved using the Sinkhorn algorithm. Since FROT finds the transport plan from selected features, it is robust to noise features. To show the effectiveness of FROT, we propose using the FROT algorithm for the layer selection problem in deep neural networks for semantic correspondence. By conducting synthetic and benchmark experiments, we demonstrate that the proposed method can find a strong correspondence by determining important layers. We show that the FROT algorithm achieves state-of-the-art performance in real-world semantic correspondence datasets.

Estimating mutual information is an important statistics and machine learning problem. To estimate the mutual information from data, a common practice is preparing a set of paired samples $\{(\mathbf{x}_i,\mathbf{y}_i)\}_{i=1}^n \stackrel{\mathrm{i.i.d.}}{\sim} p(\mathbf{x},\mathbf{y})$. However, in many situations, it is difficult to obtain a large number of data pairs. To address this problem, we propose the semi-supervised Squared-loss Mutual Information (SMI) estimation method using a small number of paired samples and the available unpaired ones. We first represent SMI through the density ratio function, where the expectation is approximated by the samples from marginals and its assignment parameters. The objective is formulated using the optimal transport problem and quadratic programming. Then, we introduce the Least-Squares Mutual Information with Sinkhorn (LSMI-Sinkhorn) algorithm for efficient optimization. Through experiments, we first demonstrate that the proposed method can estimate the SMI without a large number of paired samples. Then, we show the effectiveness of the proposed LSMI-Sinkhorn algorithm on various types of machine learning problems such as image matching and photo album summarization. Code can be found at https://github.com/csyanbin/LSMI-Sinkhorn., ECML/PKDD 2021

With the aim of enhancing both reliability and fatigue life of gasket, this study combines finite element analysis, orthogonal experimental design, dynamically-guided multi-objective optimization, and the non-dominated sorting genetic algorithm with elitist strategy to optimize the geometric parameters of the cylinder gasket. The finite element method was used to analyze the temperature field, thermal-mechanical coupling stress field, and deformation of cylinder gasket. The calculation results were experimentally validated by measured temperature data, and comparison results show that the maximum error between calculated value and experiment value is 7.1%, which is acceptable in engineering problems. Based on above results and orthogonal experiment design method, the effects of five factors, including diameter of combustion chamber circle, diameter of coolant flow hole, length of the insulation zone between third and fourth cylinders, thickness of gasket, and bolt preload, on three indexes: temperature, stress, and deformation of gasket, were examined in depth. Through the variance analysis of the results, three important factors were identified to proceed later calculation. The dynamically guided multi-objective optimization strategy and the non-dominated sorting genetic algorithm were effectively used and combined to determine the optimal geometric parameters of cylinder gasket. Furthermore, calculation results suggest that temperature, stress, and deformation of the optimized cylinder gasket have been improved by 27.88 K, 16.84 MPa, and 0.0542 mm, respectively when compared with the origin object, which shows the excellent performance of gasket optimization and effectiveness of the proposed optimization strategy.

Background We present a study of the etiological factors and clinical features of classic fever of unknown origins (FUO) to provide evidence for related clinical diagnosis and treatment. Methods We analyzed 1,641 cases of patients with classic FUO hospitalized in West China Hospital of Sichuan University during January 1, 2011 and December 31, 2017, analyzed the etiological factors of classic FUO, explored and screened the laboratory indicators related to infectious diseases, and compared the above data with the clinical features of tuberculosis and lymphoma, which are difficult to diagnose. Results 1,504 patients were diagnosed through various examination or diagnostic treatment methods, and the diagnosis rate was 91.65%. 48.69% (799) were infectious diseases, of which tuberculosis was the most common, accounting for 19.50% (320) . 19.26% (316) were connective tissue diseases, among which adult-onset Still’s disease (AOSD) was the most common, accounting for 5.42%(89). 16.94% (278) were neoplastic diseases, and lymphoma was the most common malignant tumor, accounting for 8.71%(143). 6.76% (111) were other diseases; and 8.35% (137) were unclear. We found that there were no significant differences between the symptoms, signs, and non-specific routine examination results of tuberculosis and lymphoma. Tuberculosis-related examinations and pathological examinations used more frequently. Conclusion The causes of classic FUO are mostly infectious diseases, in which tuberculosis accounts for a large proportion. Non-infectious diseases that cause FUO are mainly connective tissue diseases and malignant tumors. Tuberculosis and lymphoma are relatively difficult to diagnose. The causes of most cases of classic FUO could be ascertained.

A meta-model is trained on a distribution of similar tasks such that it learns an algorithm that can quickly adapt to a novel task with only a handful of labeled examples. Most of current meta-learning methods assume that the meta-training set consists of relevant tasks sampled from a single distribution. In practice, however, a new task is often out of the task distribution, yielding a performance degradation. One way to tackle this problem is to construct an ensemble of meta-learners such that each meta-learner is trained on different task distribution. In this paper we present a method for constructing a mixture of meta-learners (MxML), where mixing parameters are determined by the weight prediction network (WPN) optimized to improve the few-shot classification performance. Experiments on various datasets demonstrate that MxML significantly outperforms state-of-the-art meta-learners, or their naive ensemble in the case of out-of-distribution as well as in-distribution tasks., 12 pages

The emphasis of this paper lies in the development of an efficient approach to reproduce the behaviors of the scramjet-powered hypersonic system with high fidelity. The modeling of the dual-mode scramjet powered hypersonic vehicle dynamics with shock interaction, Ram-to-Scram transition, and finite rate chemistry reaction is firstly introduced. The structure of surrogate model is identified with the implement of iterative fractional factorial design (IFFD). In order to declare the reliability of the surrogate models, ν-gap metric is applied to distinguish the difference among these surrogate models in terms of closed-loop performance. The results show that the influence of Mach number on the aerodynamics should not be overlooked, and the effect of propulsion system to the aerodynamic pitch moment is dramatic. Further, the partial Kriging model appears to have the closest plants throughout the flight envelope compared with the full Kriging model and polynomials model. Nevertheless, considering the briefness of analytical expression, the polynomials model may be an alternative approach for design-oriented modeling.

This paper has researched the influence on body shape design of hypersonic vehicle on flight performance and control stability. Control-oriented parametric model is established, and some important geometric parameters of the parametric model are put into the expression of aerodynamic coefficient to measure the degree of influence on performance. The results show that the length of the elevator influences the control effect and the upper angle of forebody effects on the balanced force and moment significantly. Consequently, the geometric parameters are easily determined in model design and the number of iterations between body system and control design is reduced.

This paper studies the conditions of the successful takeoff and climbing of the aerospace vehicle. By analyzing the longitudinal kinematics model and dynamic model of the vehicle, the take-off constraints determined by integrating with the lift-drag ratio, thrust-to-weight ratio, flight speed and take-off weight. Moreover, the algebraic form of the maximum take-off weight is also derived to ensure that the aircraft can successfully take off under a fixed configuration. Finally, the simulation results verify the effectiveness of the proposed methods.

In traditional networks, interfaces of network nodes are duplex. But, emerging communication technologies such as visible light communication, millimeter-wave communications, can only provide a unidirectional interface when cost is limited. It's urgent to find effective solutions to utilize such new unidirectional communication skills. Decoupling implies separating one single resource to two independent resources. This idea can be applied at physical layer, link layer, network layer, even transport layer. TCP decoupling is an end to end solution provided at transport layer. With decoupled TCP, two distinct unidirectional path can be created to meet the requirements of reliable information transfer. However, it is not an easy task to decouple a bidirectional logical path at transport layer. In this paper, we dwell on the idea of TCP decoupling. Advantages of decoupling at transport layer are analyzed also. In addition, an experiment is carried out to figure out how to implement a decouple TCP. Our results show decoupling at transport layer is possible and the modified protocol is available., 9 pages, 12 figures

In order to reduce the maintenance cost, the design of complex electronic devices need to consider Failure Modes and Effects Analysis (FMEA). Traditional FMEA method in face of complex devices is usually lack of a formal specification in analysis process of failure modes and causes, which is easy to cause confusion and chaos; the results are prone to duplication or omission; the support for design phase is also weak. Above all, this paper presents a FMEA method for electronic devices based on annotating functional roles. Firstly, the functional roles and their corresponding failure modes are introduced; secondly, possible failures, that each functional role may experience, are analyzed and a formalized analysis method based on stress event is proposed; finally, analysis process of the FMEA method is illustrated with avionics system battery as an example. The method presented in this paper can effectively improve the efficiency and accuracy of failure analysis for electronic devices, and improve the ability of FMEA to support the design phase.

In this paper, we present our solution to Google YouTube-8M Video Classification Challenge 2017. We leveraged both video-level and frame-level features in the submission. For video-level classification, we simply used a 200-mixture Mixture of Experts (MoE) layer, which achieves GAP 0.802 on the validation set with a single model. For frame-level classification, we utilized several variants of recurrent neural networks, sequence aggregation with attention mechanism and 1D convolutional models. We achieved GAP 0.8408 on the private testing set with the ensemble model. The source code of our models can be found in \url{https://github.com/ffmpbgrnn/yt8m}., CVPR'17 Workshop on YouTube-8M

Background: Percutaneous endoscopic interlaminar discectomy (PEID), which poses advantages for certain types of herniated disc, is gaining wider acceptance in clinical practice. We retrospectively analyzed the efficacy of the PEID technique in treatment of calcified lumbar disc herniation. Study Design: A retrospective case-control study. Setting: University hospital in China. Objective: To evaluate the efficacy of the PEID technique in treatment of calcified lumbar disc herniation, and a comparison between calcified and noncalcified disc herniation was drawn to analyze the causes of herniated disc calcification. Methods: Data from patients who underwent full-endoscopic lumbar discectomy in our department between March 2011 and May 2013 were collected. Thirty cases with calcified lumbar disc herniation were included in the study group, and 30 age-, gender-, and body mass index (BMI)-matched cases with noncalcified lumbar disc herniation served as controls. Perioperative data, preoperative and postoperative Visual Analog Scale (VAS) scores, Oswestry Disability Index (ODI) values, MacNab scores, and postoperative low-extremity dysesthesia among patients in the 2 groups were collected. Results: The values of computed tomography (CT) in the calcified group were significantly higher than those in the noncalcified group (P < 0.01). The preoperative disease courses in the 2 groups were similar. However, there was a statistically significant difference in the duration of traditional Chinese medicines (TCM) administration (P < 0.01). VAS and ODI scores improved significantly after surgery, but there were no significant differences between the 2 groups (P > 0.05). Three months after surgery, the rate of low-extremity dysesthesia in the calcified group was significantly higher than that in the control group (P = 0.03) but became similar at 6 months. By applying MacNab criteria the proportions of good and excellent were greater than 90% in both groups, and there was no difference between groups (P > 0.05). Limitations: The sample size was small in this retrospective study. Conclusion: The PEID technique is an effective method in the treatment of calcified lumber disc herniation, although the rate of postoperative dysesthesia is higher in this group during the early postoperative period. Long-term TCM administration may be related to the calcification of herniated lumbar discs. Key words: Lumbar disc herniation, percutaneous endoscopic lumbar discectomy, interlaminar approach, calcification

Deep ConvNets have shown its good performance in image classification tasks. However it still remains as a problem in deep video representation for action recognition. The problem comes from two aspects: on one hand, current video ConvNets are relatively shallow compared with image ConvNets, which limits its capability of capturing the complex video action information; on the other hand, temporal information of videos is not properly utilized to pool and encode the video sequences. Towards these issues, in this paper, we utilize two state-of-the-art ConvNets, i.e., the very deep spatial net (VGGNet) and the temporal net from Two-Stream ConvNets, for action representation. The convolutional layers and the proposed new layer, called frame-diff layer, are extracted and pooled with two temporal pooling strategy: Trajectory pooling and line pooling. The pooled local descriptors are then encoded with VLAD to form the video representations. In order to verify the effectiveness of the proposed framework, we conduct experiments on UCF101 and HMDB51 datasets. It achieves the accuracy of 93.78\% on UCF101 which is the state-of-the-art and the accuracy of 65.62\% on HMDB51 which is comparable to the state-of-the-art.

In the light of the broad range of hypersonic flight conditions, a design method of switching control based on guardian maps is proposed to guarantee the global stability within the flight envelope. First, a linear parameter varying (LPV) model of a hypersonic vehicle is established over a range of hypersonic flight conditions. The parameters of controller are determined by pole placement technology with the target pole locations at the boundaries of the parameters variation range. Second, a new range of flight conditions that ensures system stability with current controller is analyzed based on guardian maps. The execution of the iterative process is used to determinate the set of controller parameters overall the flight envelope with initial target pole locations. By switching controller parameters the closed-loop performance is guaranteed over the entire flight envelope. The simulation shows that the proposed design method has the ability to ensure the global stability over the flight envelope, and achieving good tracking performance.

Compared with traditional vehicles, the dynamic model of hypersonic vehicles features higher nonlinearity, tighter coupling and stronger uncertainty. What's more, the designed controller needs to have more stringent performance demands as well as robustness requirements. According to the guardian maps theory, a proposed design method can adaptively adjust control parameters in relation to the given control structure for hypersonic vehicles. First, a linear parameter varying (LPV) model of a hypersonic vehicle is established, and after that handling qualities are translated in terms of eigenvalue confinement. Then for a fixed controller architecture and initial controller gains, a set of control parameters is adjusted automatically using this theory while ensuring generalized stability in the expected flight process. Finally, an example of a hypersonic vehicle is provided in simulation to verify the feasibility of the proposed control method.

The turbine system of a two-stage turbocharger composed of high pressure turbine (HT), low pressure turbine (LT), and by-pass valve decides distribution and utilization of exhaust gas energy and influence performance of two-stage turbocharger in whole operational conditions. Besides, characteristics of turbine is expressed by envelop line of characteristic lines in different speeds. So turbine can be conveniently selected compared with compressor with similarity theory. Therefore, two-stage turbocharger matching begins from turbine system matching in the paper. In two-stage turbocharger, cooler efficiency, cooler loss, and by-pass valve open besides turbochargers will influence turbocharging system performance and design of cooler and by-pass valve are important contents of turbocharging system matching. The paper matched intercooler, by-pass valve open, compressors, and turbines jointly. Calculation model for turbocharger matching was built, and turbine performance is get from reference turbine based on similarity theory; influence of compressor ratio distribution, cooler efficiency, and pressure drop in cooler imposing on compressor work was analyzed; and influence of turbine flow capacity and by-pass valve imposing on output working in expanding process was studied; the method for matching of two-stage turbocharging system in whole operational condition is studied. Matching analysis was made aiming at two-stage turbocharging system of a type of high power density diesel engine, and design for turbocharging system was finished. Matching result using the method is compared to matching result using traditional method. Analysis result proves that using the method matching points in different operational conditions are located in more reasonable zone of compressor MAP.

The objective of this study was to compare the efficacy and safety of piperacillin-sulbactam (PIP-SBT) and piperacillin-tazobactam (PIP-TAZ) in the treatment of bacterial respiratory and urinary tract infections. A randomised, single-blind, controlled clinical trial was performed. Differences in clinical efficacy, bacteriology and safety between the two groups were subjected to statistical analysis, including intent-to-treat (ITT) analysis. A total of 215 cases were enrolled, with 203 complete cases (99 PIP-SBT, 104 PIP-TAZ). A total of 209 cases (103 PIP-SBT, 106 PIP-TAZ) were included in the ITT analysis and a total of 212 cases (104 PIP-SBT, 108 PIP-TAZ) were included in the safety analysis. Overall efficacy rates of PIP-SBT and PIP-TAZ were 93.2% and 93.4%, respectively. Overall bacterial eradication rates of the two groups were 95% and 97.59%, respectively. Among the PIP-SBT group, eight patients (7.69%) had adverse events, including four probable drug-related events. Among the PIP-TAZ group, nine patients (8.33%) had adverse events, including one definitely drug-related and four probable drug-related events. All differences between the two groups were insignificant. PIP-SBT could be a suitable replacement for PIP-TAZ in the therapy of community-acquired respiratory and urinary tract infections caused by beta-lactamase-producing bacterial isolates.

Sir, We read the article by Qu et al1 with great interest. The authors investigated the role of procalcitonin (PCT) in distinguishing bacterial infections in febrile patients. We believe that this topic can be enriched with the points listed below: Groups 1 and 2 represented bacterial infection groups which consisted of 161 hospital acquired infection cases (58%) and 115 community acquired infection cases (42%). If there are significant differences in the levels of PCT, C-reactive protein (CRP), interleukin-6 (IL-6) and serum amyloid A (SAA) between hospital acquired infection and community acquired infection cases, it can be useful in guiding antibiotic treatment. Bacterial pneumonia was the most common diagnosis in their study, but Streptococcus pneumoniae which is the most common aetiological agent of pneumonia, was not isolated in any of their febrile patients. This point should be clarified. The PCT level of some patients in the bacterial infection group was below the cut-off level. It is important to know which infections and bacteria cause low PCT levels. Mortality among patients was not mentioned. It would be important to know whether there was a significant difference in PCT levels between survivors and non-survivors. Finally, we think that the aforementioned data will be helpful to increase the significance of PCT especially in emergency department and intensive care unit.

The study of digital Chinese calligraphy has become more valuable nowadays. While existing researches on Chinese calligraphy analysis are primarily focused on stroke-based character recognition and simulation, we propose a global feature descriptor to deal with style recognition problem in this paper. The proposed method extracts three categories of character features: position features, proportion features and projection features. These features are then used to train an SVM classifier of calligraphy style. We test the global feature based classifier on five-style Chinese calligraphy character set. The experimental results show that the proposed method can achieve good classification accuracy, proving the effectiveness of the global feature descriptor in calligraphy style recognition.

This paper is concerned with the convergence, global superconvergence, local superconvergence, and stability of collocation methods for ${u}^{\prime }(t)=au(t)+bu([t])$ . The optimal convergence order and superconvergence order are obtained, and the stability regions for the collocation methods are determined. The conditions that the analytic stability region is contained in the numerical stability region are obtained, and some numerical experiments are given.

Market information management has always been an essential component of good marketing management practice, but only possession of market information is not sufficient. Recently, processing market information has been of increasing interest to organization researchers, they suggested that improving the market information processing capabilities is important for marketing managers to understand and track changes in users’ needs, wants, preferences and consumption patterns. However, although the literature pertaining to the study of organizational information processing is growing, there lacks a specific analysis as to using online users’ information behavior to process market information, research is still needed to guide firms on how to integrate and manage market information based on consumer behavior. In an attempt to fill in the gap, this study reviews the relevant literatures and proposes that using users’ three-dimensional information behavior to capture the user’s interest in information and filter incoming information will be a very efficient way to process market information. We will use the Mobile Internet click stream data from some WAP website to characterize the users’ online information behavior. In addition, we will examine the relationships between the three dimensions of consumer online information behavior and the performance of the website. This study attempts to contribute to a better understanding of market information processing, our understanding of how organizations process market information can be advanced substantially on the basis of principles information processing.

Nowadays, the competition among enterprises has transformed into the competition within supply chains under the circumstance of rapidly developing E-commerce age, the reduction of logistics cost has already become the core element of the enterprises competition. This paper aims to establish SCOR-based enterprises ABM model, which includes logistics cost calculation, performance evaluating determination and process improvement and so on. By the study of ABM model and the analysis of the activities costing, we can improve the process and then enhance the competitiveness of the enterprises.

For complicated Software-Intensive System, it is always hard to guarantee the reliability and safety of software. Effective methods for detecting faults and isolating software fault from hardware fault are desiderated especially. In order to detect and isolate fault in SIS, a method called runtime software monitoring is studied, and a new kind of runtime software monitoring system(RSMS) is constructed in this paper. The RSMS can not only detect software fault by observing software behavior to determine whether it complies with its intended behavior, but also can assist to isolate software fault from hardware fault and to locate software fault based on fault symptoms acquired by our method. The software architecture of RSMS is presented from different views by using “4+1” view model and layer architectural style. The RSMS prototype is implemented through architecture-based software development method. By applying the prototype in practice, it proved that the RSMS prototype is feasible and effective for detecting and diagnosing faults in SIS.

Reputation systems are designed for reducing the risk entailed in interactions among total strangers in electronic marketplaces. Such systems can be used to collect and aggregate feedback on the past behavior of participants in electronic transactions, so as to derive reputation scores assumed to predict likely future behavior.The security of reputation system is vital for the system to run correctly, if reputation system is not secure, it may not be able to produce a correct result, and consequently affects the process of decision making. In this paper, a secure reputation system based on secure multiparty computation is proposed. Each of feedback provider generates an anonymous identity using smart card, and provides a rating about feedback target by a secure multiparty computation to a querying user. The system contains the following desirable features: privacy-preserving of feedback provider, tamper-preventing of feedback, masquerade-proofing of identity, shilling attack-preventing.

ERA is the acronym of environment, reactive rules and agent; the model is a new effective multi-agent cooperation framework, which has been successfully applied in a wide range of areas. Multi-port Container Stowage (MPCS) is a more practical problem. However, multi-port makes the container stowage problem more difficult. This paper presents a MPCS model based on time series and a multi-agent ERA model based on belief to solve the multi-port container stowage problem. The algorithm introduces the belief concept into agents and creates a multi-agent ERA model. The searching direction and strength of the agents are determined by the learning machine using belief parameter. Agent has the ability to evaluate its searching path. At last, the feasibility and efficiency of the model are validated by the experimental result.