Search

Deep learning (DL) libraries reduce the barriers to the DL model construction. In DL libraries, various building blocks are DL operators with different functionality, responsible for processing high-dimensional tensors during training and inference. Thus, the quality of operators could directly impact the quality of models. However, existing DL testing techniques mainly focus on robustness testing of trained neural network models and cannot locate DL operators’ defects. The insufficient test input and undetermined test output in operator testing have become challenging for DL library developers. In this article, we propose an approach, namely Duo, which combines fuzzing techniques and differential testing techniques to generate input and evaluate corresponding output. It implements mutation-based fuzzing to produce tensor inputs by employing nine mutation operators derived from genetic algorithms and differential testing to evaluate outputs’ correctness from multiple operator instances. Duo is implemented in a tool and used to evaluate seven operators from TensorFlow, PyTorch, MNN, and MXNet in an experiment. The result shows that Duo can expose defects of DL operators and realize multidimension evaluation for DL operators from different DL libraries.

Temperature is one of the main loads of super-high arch dams. (1) Background: a super-high arch dam in southwest China was taken as an example in this paper and the temporal-spatial evolution law of dam temperature was analyzed based on the monitoring data. (2) Methods: the finite element simulation analysis method was adopted to invert the boundary conditions of temperature on the upstream surface and the thermal parameters of the concrete, and the temperature evolution process of the arch dam in long-term operation was simulated and analyzed. After the distribution characteristics of the designed reservoir water temperature and the actual reservoir water temperature were compared, the difference in the temperature field of the arch dam during the impoundment and operation under the designed and actual conditions was studied. (3) Results: the temporal-spatial evolution law of the temperature in the dam operation period accords with the conventional knowledge, and the calculated value through simulation feedback is in good agreement with the monitoring value, which can reflect the actual temperature field distribution of the dam. (4) Conclusions: under the design condition, the dam temperature rose slowly after closure grouting and then tended to be stable. Under the actual condition, the temperature rose by 7.1~9.2 °C after closure grouting, reached the highest temperature in about 8~12 years, and fell back to a stable temperature in 40~80 years.

Замечательным методом исследования решений нелинейного солитонного уравнения является метод $\bar\partial$-одевания. Хотя для этой цели также могут применяться другие подходы, $\bar\partial$-одевание оказывается наиболее понятным и непосредственно приводящим к конечным результатам. Изучается $(2+1)$-мерное уравнение Савады-Котеры. Для его решения используются собственные функцие и функция Грина представления Лакса, а также обратное спектральное преобразование, которое трансформируется в новую $\bar\partial$-задачу. На основе этой $\bar\partial$-задачи путем выбора надлежащего спектрального преобразования строится решение. Если задана временна́я эволюция спектральных данных, то можно получить окончательное формальное выражение для решения уравнения Савады-Котеры.

The ultrasonic Lamb wave total focusing method (TFM) only uses the amplitude of the defective scattered signal for virtual focused imaging, while ignoring the phase information of the scattered signal and the dispersion characteristics of the Lamb wave, resulting in low imaging resolution and easily produced artefacts in imaging. To solve this problem, an ultrasonic Lamb wave imaging method based on phase coherence is proposed in this paper and the sign coherence factor (SCF) in the full matrix scattering signal is extracted. Moreover, the method uses the SCF to weight the amplitude of the full matrix scattering signal, suppresses the side lobes of the defect echo signal and the Lamb wave dispersion effect, improves the ultrasonic Lamb wave imaging resolution and weakens the artefacts. Finally, single- and multiplehole defects in aluminium plates are detected for experimental validation using an ultrasonic phased array. The array performance indicator and signal-to-noise ratio are used as indicators for quantitative assessment of imaging performance. The results show that compared with the TFM imaging, the SCF imaging can effectively suppress the noise and scattered signal side lobes, improve the array performance indicator (API) by 69.1% and improve the signal-to-noise ratio (SNR) by 73.9%. In addition, the SCF imaging can effectively weaken the interference of scattered signals between multiple through-hole defects, resulting in fewer artefacts in imaging.

Nisoldipine can effectively suppress pulmonary arterial smooth muscle cell proliferation and c-Fos expression. To identify the mechanism of the anti-inflammatory effects in monocrotaline-induced pulmonary arterial hypertension (PAH), focusing on the c-Fos/NLRP3/caspase-1 pathway. In a mice model of monocrotaline-induced PAH, miRNA-155 expression was increased. In an in vitro model, overexpression of miRNA-155 promoted inflammation and induced c-Fos, NLRP3, and caspase-1 protein expression. The inhibition of c-Fos reduced the effects of miRNA-155 on inflammation in an in vitro model of monocrotaline-induced PAH. The inhibition of NLRP3 reduced the effects of miRNA-155 on inflammation in an in vitro model of monocrotaline-induced PAH. miRNA-155 increased inflammation in monocrotaline-induced PAH through c-Fos/NLRP3/caspase-1.

Our general purpose was to examine the effect of condensed Fuzheng extract (CFE) on the alleviation of immunosuppression. A mouse model of immunosuppression was established by intraperitoneal injection of CTX. A healthy control group received no CTX and no CFE; different intragastric doses of CFE were administered to three groups of mice for 28 days (4500, 2250, or 1125 mg/kg/day); a negative control received CTX alone, and a positive control received CTX and levamisole hydrochloride. We evaluated the effects of CFE on the immune system organs, cells, and molecules by comparing the different groups. CFE significantly improved immune system organs (spleen and thymus indices and histology), stimulated immune cell activities (number of white blood cells and lymphocytes, phagocytosis of mononuclear phagocytes, proliferation of splenic lymphocytes, antibody formation, and NK cell activity), and increased the levels of immunoglobulins (IgA, IgG, and IgM) and cytokines (IL-2 and IFN-γ). Thus CFE effectively alleviated CTX-mediated immunosuppression and oxidative stress and enhanced the immunological functions of mice.

PTEN is known as a tumor suppressor and plays essential roles in brain development. Here, we report that PTEN in primary sensory neurons is involved in processing itch and thermal information in adult mice. Deletion of PTEN in the dorsal root ganglia (DRG) is achieved in adult Drg11-Cre

With the wide use of Deep Learning (DL) systems, academy and industry begin to pay attention to their quality. Testing is one of the major methods of quality assurance. However, existing testing techniques focus on the quality of DL models but lacks attention to the core underlying inference engines (i.e., frameworks and libraries). Inspired by the success stories of fuzz testing, we design a graph-based fuzz testing method to improve the quality of DL inference engines. This method is naturally followed by the graph structure of DL models. A novel operator-level coverage criterion based on graph theory is introduced and six different mutations are implemented to generate diversified DL models by exploring combinations of model structures, parameters, and data inputs. The Monte Carlo Tree Search (MCTS) is used to drive DL model generation without a training process. The experimental results show that the MCTS outperforms the random method in boosting operator-level coverage and detecting exceptions. Our method has discovered more than 40 different exceptions in three types of undesired behaviors: model conversion failure, inference failure, output comparison failure. The mutation strategies are useful to generate new valid test inputs, by up to an 8.2% more operator-level coverage on average and 8.6 more exceptions captured.

Interface engineering has been demonstrated to be an effective strategy for enhancing the thermoelectric (TE) performance of materials. However, a very typical interface in semiconductors, that is, the PN junction (PNJ), is scarcely adopted by the thermoelectrical community because of the coexistence of holes and electrons. Interestingly, our explorative results provide a definitively positive case that appropriate PNJs are able to enhance the TE performance of p-type Sb2Te3-based alloys. Specifically, owing to the formation of the charge-depletion layer and built-in electric field, the carrier concentration and transport can be optimized and thus the power factor is improved and the electronic thermal conductivity is decreased. Meanwhile, PNJs provide scattering centers for phonons, leading to a reduced lattice thermal conductivity. Consequently, the p-type (Bi2Te3)0.15–(Sb2Te3)0.85 composites comprising PNJs achieve a ∼131% improvement of the ZT value compared with the pure Sb2Te3. The increased ZT demo...

Scene classification of remote sensing images plays an important role in many remote sensing image applications. Training a good classifier needs a large number of training samples. The labeled sam...

It is time consuming to annotate unlabeled remote sensing images. One strategy is taking the labeled remote sensing images from another domain as training samples, and the target remote sensing labels are predicted by supervised classification. However, this may lead to negative transfer due to the distribution difference between the two domains. To address this issue, we propose a novel domain adaptation method through transferring the parameters of extreme learning machine (ELM). The core of this method is learning a transformation to map the target ELM parameters to the source, making the classifier parameters of the target domain maximally aligned with the source. Our method has several advantages which was previously unavailable within a single method: multiclass adaptation through parameter transferring, learning the final classifier and transformation simultaneously, and avoiding negative transfer. We perform experiments on three data sets that indicate improved accuracy and computational advantages compared to baseline approaches.

This paper aims to adopt the total focusing method (TFM) and wavenumber method for characterizing a delamination in ballastless slab track. Twelve dry point contact (DPC) transducers located at the upper surface of the slab track compose a linear array. These transducers are employed to actuate shear waves, which are suitable for identifying the delamination. The technique of removing the surface wave has been implemented for only retaining the scattered wave caused by the delamination and the reflected wave from the bottom of bed plate. Numerical and experimental results demonstrate that the delamination and bottom of the bed plate can be identified by the proposed methods. Furthermore, the near-surface pseudomorphism is distinctly restrained after removing the surface wave. Compared to TFM, the wavenumber method has the great advantages of improving computational performance and lateral resolution. However, they have no significant difference in the longitudinal resolution. Furthermore, it has been confirmed that the lateral resolution can be affected by the amount of transducers. This paper can provide valuable suggestions on improving the computational performance and the imaging accuracy when we identify a delamination in ballastless slab track.

The method of continuous melting and casting was used in the research to attempt to acquire the ternary directionally solidified Ti-Al-Nb alloy by once casting. It was investigated about microstructure and mechanical properties of the Ti44Al2Nb alloy by changing the temperature gradient. Results show that the lamellar spacing decreased to 1.25 µm, when the temperature gradient increases and the composition is relatively uniform in different regions. The tensile strength, the compressive strength and the fracture toughness of Ti44Al2Nb alloy is 455 MPa, 1770 MPa and 23.5 MPa m1/2 respectively, under the parameters of 50 kW and 0.5 mm/s. The high temperature compressive strength increases, when the strain rate increases and temperature decreases. Because the large strain rate shortens the strain time and piles up dislocations. Another, the low temperature provides the less driving force of dislocation movement. To observe the crack propagation process and the fracture morphology, the fracture mode contains trans-lamellar fracture and interface de-lamination. The method is a feasibility way to prepare the directionally solidified TiAlNb alloy and to add the high-melting-point element into microstructure directly.

A multi-objective particle swarm optimization algorithm, based on share–learning and dynamic crowding distance (MOPSO-SDCD), is proposed to improve the convergence accuracy and keep the diversity of the Pareto optimal solutions. First, the sharing–learning factor is applied to modify the velocity updating formulas, which improves both the global search ability and local search accuracy of the algorithm. Meanwhile, Gaussian mutation and greedy strategy are adopted to update personal best position and external archive, which make the algorithm approximate the Pareto front quickly and avoid premature convergence. Finally, MOPSO-SDCD maintains the external archive based on dynamic crowding distance sorting strategy, whose purpose is boosting the diversity and distribution of Pareto optimal solutions. The ZDT series test functions are used to test the performance of MOPSO-SDCD and compare with other three typical algorithms. Simulation results verify the superiority and effectiveness of the proposed algorithm.

For improving the convergence accuracy and diversity of multi-objective optimization algorithm a multi-objective quantum-behaved particle swarm optimization algorithm with double-potential well and share-learning is proposed, which overcomes the deficiency of particles readily gathering in identical solutions. The two local attractors, inside and outside, are introduced to construct the particle locations updating model, using the quantum tunneling and transition effects in double-potential well model. In this way, the particle moves to the solution sparseness region in later evolution stage, so as to avoid gathering in the single local attractor and escape from local optimum. Therefore the optimization accuracy of the algorithm is improved. The share-learning strategy is adopted to extend the search range of particles and increase the diversity of solutions. The problem of easily converging to boundary solutions in quantum-behaved particle swarm optimization algorithm could be avoided. Simulation results show that the proposed algorithm makes excellent performance in optimization accuracy, convergence, diversity, and distribution, compared with three existing algorithms. Moreover, the proposed algorithm can hold on better convergence and distribution performance when handling high-dimensional multi-objective problems.

The error spectrum is a comprehensive metric for evaluation of estimation performance, which is an aggregation of many incomprehensive measures. However, for dynamic systems, error spectrum is difficult to analyze an estimator's performance since it is a three-dimension graphics for the whole time horizon. Accordingly, dynamic error spectrum was proposed to relieve error spectrum to dynamic systems. The key of dynamic error spectrum is combining error spectrum into one point at a time instant, which unfortunately suffers from information loss. To improve this, a new metric called enhanced dynamic error spectrum is proposed in this paper. Then, both the additive form and the multiplicative form of enhanced dynamic error spectrum are given under different application backgrounds. Finally, the concrete application of the metrics and qualities of the above form are illustrated by numerical examples. The simulation shows that more important information are considered by enhanced dynamic error spectrum and the proposed measure can also be applied easily to dynamic systems.

Binary Bell polynomial approach is applied to study the coupled Ramani equation, which is the generalization of the Ramani equation. Based on the concept of scale invariance, the coupled Ramani equation is written in terms of binary Bell polynomials of two dimensionless field variables, which leads to the bilinear coupled Ramani equation directly. As a consequence, the bilinear Bäcklund transformation, Lax pair and conservation laws are systematically constructed by virtue of binary Bell polynomials.

Interface engineering has been demonstrated to be an effective strategy for enhancing the thermoelectric (TE) performance of materials. However, a very typical interface in semiconductors, that is, the PN junction (PNJ), is scarcely adopted by the thermoelectrical community because of the coexistence of holes and electrons. Interestingly, our explorative results provide a definitively positive case that appropriate PNJs are able to enhance the TE performance of p-type Sb

Our aim was to investigate the role of interleukin-6 (IL-6) in myocardial regeneration from mice after cardiac injury. The newborn mice were divided into the following 4 groups (16 in each group): sham group, model group, IL-6-/- group (IL-6 knockout) and IL-6 group (IL-6 overexpression). Electrocardiography was performed on all mice and found higher LVEDD, LVESD and IVST and lower LVEF and LVFS in the IL-6 group compared with the sham group. Using HE staining, severe myocardial injury combined with infarction and fibrosis were observed in the IL-6-/- group, while the damaged myocardial tissue was repaired to some extent in the IL-6 group. The expression of IL-6 in the IL-6 group were significantly up-regulated. BrdU immunofluorescence found that the IL-6-/- group had the least number of BrdU positive cells, while the IL-6 group had more BrdU positive cells than the model group and the IL-6-/- group. Expressions of IL-6, cyclinD1 and Bcl-2 in the IL-6 group were up-regulated compared with other groups. In conclusion, IL-6 overexpression could enhance cardiomyocyte proliferation and relevant protein expression in mice myocardium, thus promoting cardiac regeneration.

As long-standing clinical problems, catheter-related infections and other chronic biofilm infections are more difficult to treat due to the high antibiotic resistance of biofilm. Therefore, new treatments are needed for more effective bacteria clearance. In this study, we evaluated the antibacterial activities of several common antibiotics alone and their combinations against biofilm-embedded methicillin-resistantstaphylococcus aureus(MRSA) infections, bothin vitroandin vivo. In brief, fosfomycin, levofloxacin, and rifampin alone or in combination with linezolid were testedin vitroagainst planktonic and biofilm-embedded MRSA infection in three MRSA stains. The synergistic effects between linezolid and the other three antibiotics were assessed by fractional inhibitory concentration index (FICI) and time-kill curves, where the combination of linezolid plus fosfomycin showed the best synergistic effect in all strains. For further evaluationin vivo, we applied the combination of linezolid and fosfomycin in a catheter-related biofilm rat model and found that viable bacteria counts in biofilm were significantly reduced after treatment (P<0.05). In summary, we have shown here that the combination of linezolid and fosfomycin treatment had improved therapeutic effects on biofilm-embedded MRSA infection bothin vitroandin vivo, which provided important basis for new clinical therapy development.

Phagocytic clearance and inefficient targeting are two major concerns for nanomedicines in cancer therapy. In this study, cell membrane inspired multifunctional copolymers (PMNCFs) were synthesized by a combination of cell membrane stealthy hydrophilic phosphorylcholine (PC), hydrophobic cholesterol (Chol) and tumor targeting folic acid (FA) functionalities on the different side chain ends. PMNCF micelles were prepared in aqueous solution to form a cell membrane mimetic structure with linked folic acid ligands as the protruding antennae on the surface of the micelles. Coumarin-6 loaded PMNCF micelles indicated that the mouse peritoneal macrophage cell uptake efficiency was suppressed to 1/10 compared with that of PLA nanoparticles. Doxorubicin loaded micelle measurements demonstrated that up to 30% of the drug could be obtained forming a stable formulation under both storage and physiological conditions. Tumor cell uptake and toxicity studies revealed that FA-decorated PMNCF micelles could increase MADB-106 cell uptake by 4-fold, and DOX loaded PMNCF micelles could kill tumor cells more efficiently than the same amount of free DOX. These exciting results confirmed the great potential of the stable, stealthy and tumor cell targeting PMNCF micelles for developing advanced long circulation and target-selective drug delivery nanoparticles.

To improve the range and penetration ability, a boost-skipping trajectory is proposed for the air-breathing hypersonic missile, with its scramjet ignited in break cycle mode. The longitudinal plane motion equations are established including the thrust term of lateral jets, and constraints on path parameters and terminal conditions are modeled. Using hp -adaptive Gauss pseudospectral method, the optimal control problem is transformed into a nonlinear programming problem with a series of algebraic constraints. And then adopting the sequential quadratic programming algorithm, the optimal trajectory is obtained satisfying the constraints. Several simulations and comparative cases are carried out, simulation results illustrate the proposed hp -adaptive pseudospectral method is superior to conventional optimization method on both computing efficiency and accuracy. The range of the missile is significantly enhanced in this optimized boost-skipping trajectory, which is 3.81 times longer compared with the existing level flight trajectory.

Numerical simulations are carried out to investigate lateral jet interaction effects for the hypersonic quasi-waverider vehicle. A reasonable jet nozzles layout is proposed. Based on three-dimensional viscous compressible Reynolds Averaged Navier-Stokes codes, comparisons of the jet interaction flow structures between quasi-waverider vehicle and general axisymmetric vehicle are conducted. Furthermore, the effects of lateral jet control are examined for different angles of attack and free-stream Mach numbers. From detailed flow structure and pressure distribution analysis, the quasi-waverider vehicle gets a better effect of lateral jet without “wrap-around” the body. With the increase in angle of attack or free-stream Mach number, the attitude control moment reduces and the effect of lateral jet decreases.

In this paper, an EMD de-noising algorithm is proposed based on the statistical feature of random noise, which can eliminate the noise impaction digital integrator generated by the collected railway line state detection signals using strap-down inertial technology. Firstly, the first IMF component of the noise-dominant modes treated by the process “random sort-sum-average-reconstruc-tion”, the signal-to-noise ratio is improved while the noise power is weakened in this process. Then the signal-to-noise cut-off can be determined according to the characters of noise autocorrelation function. Finally, the global threshold could be selected by the noise-dominant mode component, so as to realize the function of filtering. The simulation and validation based on the collected railway line acceleration data using the EMD de-noising algorithm show that the noise in railway line state acceleration detection signals can be effectively eliminated using this method.

Aim: To assess the function of Ngg1 in Candida albicans and reveal the role of NGG1 in the morphological conversion and virulence of C. albicans. Materials & methods: C. albicans NGG1 gene was deleted in the wild-type strain SC5314 and the function of Ngg1 was assessed by western blot analysis. The phenotypes and the virulence of the ngg1 mutants were examined. Microarray analysis was performed to explore the mechanism. Results: The ngg1 mutants attenuated acetylated histone H3, obviously reduced filamentous growth and showed significantly diminished pathogenicity in all the infection models. Conclusion: This study suggested the histone acetyltransferase activity of C. albicans Ngg1 and revealed the important role of NGG1 in morphological conversion and virulence of C. albicans. [Formula: see text]