A deep learning⁃based semi⁃supervised prediction method for the P53 mutation status of glioma is proposed to address the current problems of poor utilisation of glioma image data and incomplete feature extraction. Firstly，NUGMB (Non⁃Uniform Granularity Multi⁃Batch) grey level partitioning algorithm is proposed to optimize the preprocessing methods of glioma MR image. Secondly，the K⁃means clustering algorithm of MCC (Multi Center Collaboration) is proposed for pseudo⁃labeling of glioma image data. Finally，a novel attention mechanism，LWAM (Local Longer and Wider Attention Modules)，is proposed to construct an improved MaxViT model based on LWAM for the preoperative non⁃invasive prediction of the P53 mutation status of glioma. The NML⁃MaxViT model based on NUGMB，MCC and LWAM algorithms predicts the P53 mutation status of glioma with an accuracy of 96.23%，which achieves non⁃invasive predictions to assist physicians in clinical diagnosis and treatment.

Single⁃cell RNA sequencing (scRNA⁃seq) technology enables researchers to measure gene expression across the transcriptome at single⁃cell resolution，progressively transforming our understanding of cell biology and human diseases. However，the high variability，sparsity，and dimensionality of single⁃cell sequencing data have significantly impeded downstream analysis，making dimensionality reduction crucial for the visualization and the subsequent analysis of high⁃dimensional scRNA⁃seq data. Yet，existing single⁃cell dimensionality reduction algorithms have not adequately considered relationships intercellular，nor have jointly optimized the tasks of dimensionality reduction and clustering. To overcome these limitations，this study focuses on scRNA⁃seq data and employs machine learning techniques to investigate a dimensionality reduction algorithm based on autoencoders. In light of the fact that most existing dimensionality reduction algorithms do not consider the use of pseudo⁃labels to supervise the training process of the encoder，leading to the loss of intercellular signals during the dimensionality reduction of data，this paper proposes a cell dimensionality reduction algorithm based on the classified autoencoder. The algorithm combines the classified autoencoder with deep embedded clustering to generate a low⁃dimensional representation of the gene expression matrix. Experimental results demonstrate that compared to six other benchmark testing algorithms，this algorithm exhibits competitive performance in a range of downstream scRNA⁃seq analysis tasks.

Superconducting quantum circuit is one of the leading approaches for realizing universal quantum computation. As the chip integration continues to increase，circuit design faces the challenges of complex wiring and signal crosstalk. In order to suppress the signal crosstalk on the chip，a novel on⁃chip structure covering coplanar waveguides，called tunnel bridge，is proposed based on the electromagnetic shielding principle of the coaxial line. In this paper，we simulate the chip circuit，with the tunnel bridge added，by using the finite element method，and the optimal impedance matching design parameters are obtained. The simulation results of the chip signal crosstalk show that，compared with conventional coplanar waveguides，the addition of the tunnel bridge brings three orders of magnitude shielding effect on the vertical electric field between top and bottom chips in flip⁃chip design. For planar single chip，the crosstalk between microwave drive lines are reduced by approximately 16 dB，and the crosstalk current caused by the DC flux bias lines are reduced by around 70%，yielding a 40% improvement over traditional air bridge schemes. The micro⁃fabrication process of tunnel bridge is also studied in this paper，and a stable sample fabrication process is achieved. The tunnel bridge structure has remarkable crosstalk suppression effect，thus a great prospect of application in large⁃scale superconducting qubit chips.

This paper proposes a parallel multi⁃path transmission design based on the CFDP (CCSDS File Delivery Protocol)，which expands the parallel transmission of files over multiple paths in space network file transfers. This design further enhances the efficiency of file transfer while maintaining compatibility with the original file transfer protocol. The design fully utilizes the characteristics of space network links and achieves the goal of parallel multi⁃path transmission by using load balancing algorithms for data scheduling to allocate file data to transmission paths reasonably. The simulation results demonstrate that compared with traditional CFDP transmission，this parallel multi⁃path transmission design fully utilizes multiple transmission paths，significantly reducing the file delivery time.

Multi⁃Instance Learning (MIL) uses labeled bags composed of multiple unlabeled instances as training data. Embedding⁃based methods address bag representation issues by embedding bags into single vectors. However，existing methods often focus on individual instances and overlook the relationship between instances and bags，which compromises the representativeness of the prototypes. Additionally，the differences between positive and negative bags are not considered by single⁃angle embedding methods，resulting in weak embedding vector quality. This paper proposes the Cluster Frequency Analysis and Dual⁃Perspective Fusion Embedding for MIL (FADE). The cluster center selection technique utilizes density peak of instances to choose a certain proportion of instances from positive and negative subspaces as cluster centers. The cluster frequency analysis technique clusters instances within subspaces based on the cluster centers，calculates cluster frequency indicators，and selects high⁃frequency cluster centers to form the prototype instance set of subspaces. The dual⁃perspective fusion embedding technique utilizes the prototype instance sets from positive and negative subspaces，along with a difference embedding function，to extract information from both perspectives and fuse the two sets of information to obtain the final embedding vector. The algorithm is tested on 29 datasets and compared with seven MIL algorithms. Experimental results demonstrate that FADE achieves higher overall classification accuracy compared to the seven benchmark algorithms，particularly excelling on image datasets while performing well on text and web datasets.

P53 gene status is an important basis for precise diagnosis and treatment of glioma. To solve the problems of incomplete heterogeneous feature extraction and multiple uncertainties inherent in the current deep learning model for MRI (Magnetic Resonance Imaging)⁃based P53 gene status prediction，we propose the precise prediction model of P53 gene status for glioma，CVT⁃RegNet (Improved RegNet integrating CNN，Vision Transfomer，and Truth Discovery). First，the RegNet network is adopted as the infrastructure of the P53 gene mutation status prediction model，which is adaptively designed to search for the heterogeneous features of the P53 gene. Second，the ViT (Vision Transfomer) module and the CNN (Convolutional Neural Networks) module are fused in the model to improve the RegNet network and further optimize the feature extraction performance and computational efficiency of the model. Finally，the Truth Discovery algorithm is incorporated for iterative optimization to improve the uncertainty of the model output，thus improving the accuracy of the prediction results. The experimental results show that the CVT⁃RegNet model predicts the P53 mutation status with an accuracy of 95.06% and an AUC score of 0.9492，which is better than the existing P53 gene status prediction models. CVT⁃RegNet realizes the non⁃invasive prediction of glioma P53 gene status，and reduces the economic burden and physical and psychological harm to patients，which provides a significant value for the precise clinical diagnosis and treatment of glioma.

The "21.7" extreme rainstorm in Zhengzhou，Henan Province，was a severe meteorological disaster that has occurred in China in recent years. The numerical models show significant uncertainty in this rainfall event，and there are deviations in the forecast of rainfall areas and intensity. Currently，the formation mechanism of the "21.7" Henan rainstorm has been widely studied，but research on its ensemble sensitivity analysis is very limited. Ensemble sensitivity analysis is a method that utilizes ensemble forecasts to estimate the sensitivity of model forecasts to initial conditions. It diagnoses the influencing factors of extreme weather processes and analyze the uncertainty of ensemble forecasts. Therefore，this study focuses on the individual case of the "21.7" Henan rainstorm，using the WRF?ARW model，combined with ensemble initial conditions，multi?physics，and model perturbations to construct serveral regional model ensemble forecasts. Ensemble sensitivity analysis is used to assess the predictability of the "21.7" Henan rainstorm and analyze the factors influencing this rainfall. The results show that the "21.7" Henan rainstorm is sensitive to the temperature field，humidity field，wind field，and geopotential height field perturbations of the initial conditions. Enhancing the cyclonic circulation in the Zhengzhou area，changing the temperature over Zhengzhou，reducing the air pressure in the Zhengzhou area，or strengthening the intensity of Typhoon In?Fa can enhance the precipitation intensity of this rainfall. This study improves understanding of the causes of the "21.7" Henan rainstorm and enhance ensemble forecasts.

Observations have revealed the presence of tornado?scale vortices (TSVs) within the boundary layer of tropical cyclones (TCs)，which significantly impact near?surface gustiness. Based on Weather Research and Forecasting Model ? Advanced Research WRF (WRF?ARW)，a high?resolution numerical experiment in simulating Hurricane Earl (2010) is conducted to analyze the activity characteristics of TSVs in boundary?layer. During the strengthening process of Earl，the number of TSVs increases with the intensification of Earl，especially during its peak. On one hand，the strengthening of the tropical cyclone (TC) provides a stronger background field，making it easier for TSVs to form. On the other hand，the stronger background field leads to an extended lifespan for TSVs，resulting in a higher number of concurrent TSVs at any given moment. The area that TSV generating gradually shifts from the left side of the vertical wind shear to the up?shear left as Earl develops，near the secondary circulation updraft and the maximum vertical vorticity，where often meets the necessary conditions of unstable vertical and horizontal wind shear. The results of TSV vorticity budget analysis further indicate that the stretching term associated with horizontal wind shear and the tilting term associated with vertical wind shear are important factors in the generation and development of TSVs，which implies that the generation and development of TSV may be related to vertical shear instability and horizontal shear instability.

Human history is also a journey of the enduring struggle against diseases. Throughout this process,medicine has evolved from mysterious witch doctors and herbal medicine to modern medicine encompassing basic medicine,clinical medicine,and public health. However,the current stage of medical development faces new challenges,particularly in translating extensive basic research findings into practical applications in clinical and public health settings. To overcome this challenge,scientists are increasingly collaborating with clinical experts to address specific clinical problems,contributing to the advancement of innovative technologies,such as novel instruments and materials,through the deepening exploration of fundamental mechanisms. This collaborative endeavor has given rise to a new interdisciplinary field within the medical domain,denoted as engineering medicine (EngMed or EM). This article briefly describes the definition,current development foundation,and key objectives of engineering medicine,and discusses future trends and challenges in this field.

The shunt loudspeaker can be used as a resonant sound absorber，as it can convert incident sound energy into electrical energy for storage and dissipation. In this paper，an analytical model is established for a rectangular rigid enclosure with shunt loudspeakers on the ceiling to analyze the noise reduction performance at low frequencies. Numerical simulations show that the sound pressure level inside the enclosure can be significantly reduced at its characteristic frequencies by using shunt loudspeakers. Better noise reduction is achieved when the shunt loudspeakers with limited numbers are placed at the locations with high initial noise level. Therefore，the top corners of the rectangle enclosure are optimal locations when employing shunt loudspeakers for broadband noise reduction，because large responses occur there for all the enclosure modes. If the ground is covered with sound absorption materials，the placement effects of the shunt loudspeakers on the noise reduction still remain although the overall sound pressure level in the enclosure decreases. Finally，experiments are conducted to verify the numerical simulation results.

3D assembly completion is an essential and complex interactive assembly task. The robot must identify the true missing parts，select the correct parts from a toolkit of candidates，calculate precise assembly poses，and ultimately make the incomplete assembly complete. As the primary principle in the design of actual assemblies such as chairs and tables，stability is also an ultimate goal of 3D assembly completion. Existing works of 3D assembly completion primarily focus on geometric relationship modeling of parts，without taking into account the stability of assembly，leading to low accuracy in completion and making it a challenge to meet the actual requirements of robot assembly. To tackle this issue，we propose StableFiT(Finishing the Incomplete 3D Assembly with Transformer) for 3D assembly completion with stability optimization. We introduce a novel stability verification method for the completed assembly. By training an assembly stability discriminator using the verification results obtained from the NVIDIA Isaac Sim simulation platform，we furtherly optimize 3D assembly completion based on stability feedback from the stability discriminator. Experimental results on the PartNet dataset demonstrate that StableFiT effectively improves the correctness and stability of the completed assemblies，addressing the limitations of existing assembly completion methods.

Gesture intention recognition is a popular research direction in the field of Human?Computer Interaction. However，most of the gesture recognition systems are based on electromyography (EMG) signals. The inevitable problems of EMG signals，such as signal crosstalk，signal attenuation，and low signal to interference plus noise ratio，seriously affect the accuracy of hand gesture recognition. To solve this problem，this paper develops the gesture intention recognition system based on liquid metal composites enabled sensor bracelet. Thanks to the sensitive piezoresistive effect of the liquid metal composites，signals obtained by the sensing bracelet exhibit excellent characteristics such as stability，high sensitivity and low noise. The gesture recognition system based on this signal consists of data acquisition and pattern recognition，and its average offline recognition accuracy is 97.19%. What's more，this system does not require additional equipment in the hand and only acquires signals through the forearm muscles. Therefore，the system can be used for disabled groups with hand function deficiency. The system has a wide range of users and important social and economic benefits.

In order to explore multi?attribute group decision?making methods in the background of interval type?2 fuzziness，the paper starts with multigranulation probabilistic rough sets and develops a multi?granularity evidence fusion decision?making model based on interval type?2 fuzzy information via with MULTIMOORA (Multi?Objective Optimization by Ratio Analysis Plus the Full Multi?Plicative Form) method and the evidence fusion theory. First，a multi?granularity interval type?2 fuzzy probabilistic rough set model is put forward. Then，the decision?maker weight and the attribute weight are calculated by the dispersion maximization method and the entropy weight method. Furthermore，in light of multigranulation probabilistic rough sets and the MULTIMOORA method，an interval type?2 fuzzy multi?attribute group decision?making model is established. The decision results are eventually acquired via the evidence fusion method from the D?S evidence theory. At last，the feasibility and effectiveness of the proposed method are verified by a case of energy consumption in steel industry. All in all，the established decision?making model in the paper owns a certain degree of fault tolerance and is conducive to acquiring stable decision results with strong interpretability.

With the rapid development of Location?Based Social Network (LBSN)，POI (Point of Interest) recommendation has gained growing attention in recent years. However，existing research still have deficiencies in mining users' POI interaction preferences and POI transferring preference，which limit their performance. To address these issues，we design an approach for personalized POI recommendation based on hybrid graph neural network (named as HGNNPR). Firstly，the method constructs the user's social network graph，and then applies the Graph Attention Networks (GAT) to learn the social representation of each user. Secondly，the method constructs labeled bipartite graph between users and POI，and then adopts the Signed Bipartite Graph Neural Networks (SBGNN) to extract users representation and POI representation including users' POI interaction preferences. Thirdly，the method constructs the directed POI transfer graph，and adopts the Session?based Graph Neural Networks （SRGNN） to learn POI representation with users' transfer preference. Then，the method integrates users' social representation and users' representation including users' POI interaction preferences to obtain the final representation of users，and integrates POI representation including users' POI interaction preferences and POI representation with users' transfer preference to get the final POI representation. Finally，final users' representation and final POI representation are combined by product operation and input into the sigmoid function to obtain the user's prediction score for POI，and then，Top?K POIs are selected and recommended to users by the sorted prediction scores. Experimental results show that，compared with the best performance of seven other methods on three datasets，our model has an average increase of 19.95% and 1.346% on the two commonly used evaluation indicators (accuracy and recall).

Currently，multi?view subspace clustering is widely studied in fields of pattern recognition and machine learning. Previous multi?view clustering algorithms mostly partition the multi?view data in their original feature space，while the efficacy of which heavily and implicitly relies on the quality of the original feature presentation. In addition，different views contain specific information in a same object and how to use these views to recover latent diverse information is particularly important for clustering.To solve the above problems，this paper proposes a method named Diversity?induced Multi?view Clustering in Latent Embedded Space (DiMCLES)，which uses projection matrix on specific view to recover latent embedded space from multi?view data. This paper uses an emprical Hibert Schmidt Independent Criterion to constrain the projection matrix on specific view which considers the diverse information of multi?view data between different views. Latent embedded learning，diversity learning，global similarity learning and clustering indicator learning are integrated into a unified framework，and an alternating optimaization scheme is introduced for optimization. Experiments on several real?world multi?view datasets verifies the superiority of our approach.

Aiming at removing irrelevant and redundant attributes，this paper proposes a method of approximate reduction to better require decision rules in intuitionistic fuzzy decision tables. The weighted system is defined via the introduction of a weighted score function in intuitionistic fuzzy sets. Additionally，variable precision rough sets are utilized for a better tolerance to misclassify. Hence，the weighted variable precision intuitionistic fuzzy sets are defined. On the basis of the constructed system，we give conceptions of the judgment theorem and identification matrix of both lower and upper approximate reduction，by which two approaches of reduction are put forward. Finally，a concrete example and numerical tests are used to illustrate the effectiveness of the proposed method.

As an important direction in research fields of artificial intelligence，granular computing has great advantages in data mining and knowledge discovery. To solve the problem of knowledge acquisition in information system with multi?scale decisions，the optimal scale selection problem of information systems with multi?scale decisions is studied by using granularity trees and cuts. The concepts of granularity trees and cuts are firstly introduced. Each attribute and decision has a granularity tree，and each granularity tree has many different local cuts，which represent the scale selection methods under a specific attribute. A local cut combination of different attributes and decision forms a global cut，resulting in a mixed scale decision table. Then，the concept of optimal cuts based on granularity trees and cuts in information systems with multi?scale decisions is presented. Finally，a comparative study between optimal cuts and optimal scale selections is performed and an algorithm is designed to verify the effectiveness of the method.

The WRF was used to numerically simulate the heavy rainfall event occurred in Northeast China from August 15 to 17 of 2019. According to the analyses of the water vapor,thermal and dynamical structures of the precipitating system,we investigated the physical mechanism of typhoon "Rosa" in enhancing the Northeast China cold vortex precipitation. The results show two main water vapor channels supplying the heavy rainfall. One is by the southwesterly airflow on the south side of the cold vortex from the Bohai Sea and the Yellow Sea while the other is by typhoon "Rosa" from the western Pacific and Japan Sea. The cold dry airmass from the inland interacts with the warm moist air from the ocean,forming an unstable stratification. Meanwhile,typhoon "Rosa" leads to an enhancement of convergence in the low?level troposphere,which produces a strong upward motion that helps release the unstable energy. This heavy rainfall event is thus caused by the coupling among these favorable water vapor,thermal and dynamic conditions.

Based on the precipitation data of 603 stations in China and the ERA5 hourly reanalysis data from 1967 to 2016,this paper analyzed characteristics of temporal and spatial variations of nocturnal precipitation in China from the perspective of nocturnal precipitation rate and nocturnal precipitation frequency. Results showed that: the stations with nocturnal precipitation were mainly distributed in the Southwest China,Northwest China,Huang Huai River Basin,Central China,Beijing?Tianjin?Hebei,northwest of South China and south of Northeast China. The nocturnal precipitation appeared most obviously in Sichuan,Tibet,Guizhou,Xinjiang and Qinghai provinces. In these areas,the nocturnal precipitation rate reached 60%,even up to 80% for individual stations,and the nocturnal precipitation frequency rate reached 55%,even up to 60% for individual stations in the whole year and all seasons. Significant seasonal variation can be seen in both spatial distribution of average nocturnal precipitation rate and nocturnal precipitation frequency rate. As for the seasonal distribution of nocturnal precipitation in China,stations with nocturnal precipitation were mainly distributed in the Southwest China,Northwest China,East China,Central China,northeast Inner Mongolia,Beijing?Tianjin?Hebei,northwest of South China and south of Northeast China in spring. During the summer,the area where this phenomenon occurred shrunk,with distribution in the Southwest China,most parts of central and southwest Xinjiang,a few areas of Gansu and Qinghai provinces,and parts of Beijing?Tianjin?Hebei,Shandong,Henan,Liaoning provinces. Almost most areas among China were covered by stations with nocturnal precipitation in autumn and winter. The paper split stations with obvious nocturnal precipitation into N?type,V?type,M?type,W?type and Λ?type respectively according to the monthly variation of average nocturnal precipitation rate and nocturnal precipitation frequency rate. These types have different characteristics of monthly variation: for N?type,the relatively high value and low value of nocturnal precipitation rate or nocturnal precipitation frequency rate commonly appeared in spring and summer respectively. Main features for V?type were relatively high value and low value of nocturnal precipitation rate or nocturnal precipitation frequency rate commonly appeared in winter and summer respectively. M?type showed significant high value of nocturnal precipitation rate or nocturnal precipitation frequency rate commonly appeared in spring and autumn,and it turned relatively low commonly in summer and winter,meanwhile W?type was classified by the relatively high value of rate commonly appeared in summer and winter and the relatively low value commonly appeared in spring and autumn. This paper also classified Λ?type by its obvious peak and valley of nocturnal precipitation rate or nocturnal precipitation frequency rate in summer and winter respectively. For the issue of distribution,N?type stations were mainly distributed in Yunnan?Guizhou Plateau,eastern foothills of the Tibetan Plateau was a typical V?type station area,while the M?type stations were mainly distributed in the Sichuan Basin. There was a consistency between monthly variation of differences between night and day of dynamic conditions and water vapor conditions affecting precipitation and the monthly variation of nocturnal precipitation rate and nocturnal precipitation frequency rate in corresponding sites,clear consistency could be found in N?type stations,which had a relatively peak value of differences between night and day of dynamic conditions in spring and relatively low value in summer. V?type region also showed this consistency in the trend of these two measures with lowest value in summer and largest in winter. Same fact appeared in M?type region with peak value in spring. Among stations with obvious nocturnal precipitation,the nocturnal precipitation rate and nocturnal precipitation frequency rate increased significantly in Nuomu Hong in Qinghai province,with nocturnal precipitation became more and more obvious. Nocturnal precipitation gradually weakened in Alashankou in Xinjiang,Haiyang in Shandong,Ruoergai and Seda in Sichuan,Nyingchi in Tibet,Nanyue in Hunan,Baoshan in Shanghai,Pingtan in Fujian and other stations,with significantly reduced nocturnal precipitation rate and nocturnal precipitation frequency rate. The nocturnal precipitation rate changed by about 1%~3% every ten years,while the nocturnal precipitation frequency rate changed by about 0.4%~1.5% every ten years. the fluctuation range of nocturnal precipitation rate with years was larger than nocturnal precipitation frequency rate in these stations.

Travel time estimation is a basic task in a smart transportation system，which is full of challenges because of its complicated spatio?temporal relationship and susceptibility to external factors. In order to obtain accurate prediction results，this paper proposes a method combining deep learning and meta?learning to predict travel time. The method is composed of a spatio?temporal network model and a meta?learning framework. The spatio?temporal network model uses convolutional neural networks and gated recurrent units to extract spatio?temporal features on the trajectory and the traffic conditions around the trajectory at the same time. The meta?learning framework is used to learn the general initialization parameters of the spatio?temporal network model from other cities，and apply the parameters to the target city. We conduct experiments on two real datasets，and the results show that the proposed method is better than the over several competitive baseline models.

Semi?supervised clustering ensemble aims at improving the accuracy of clustering ensemble by using pairwise constraints，but it achieves poor performance on high?dimensional datasets. In addition，clustering performance has little improvement when only a few pairwise constraints are available. To solve these problems，this paper proposes a novel semi?supervised clustering ensemble algorithm WSCEC (Weighted Semi?supervised Clustering ensemble algorithm based on Extended Constraint projection algorithm). Firstly，a variety of clustering algorithms are exploited to cluster the feature space of data，and then the random subspace is utilized for the dimension reduction to reduce the impact of redundant features. Secondly，the original constraint set is expanded according to the k nearest or farthest samples of constraints and the transitive relationship between constraints，and the original data space is projected into a low?dimensional space by constraint projection technique to satisfy as many constraints as possible. Finally，a weighting strategy of clustering solutions is designed，which assigns an appropriate weight to each clustering solution to reduce the impact of low?quality clustering solutions. Experimental results on several datasets prove the effectiveness of the proposed algorithm.

It is essential to acquire RNA tertiary structures for understanding and intervening their biological functions which accelerate the development of computational structure prediction. One of the key steps is to evaluate the quality of the structural candidates. Recently，approaches based on machine learning，such as AlphaFold2，have achieved revolutionary progress in protein structure prediction. In this study，we develop an RNA structure scoring function based on deep convolutional neural network. We also build a training dataset including 422 non?redundant RNAs and 126600 associated decoys. We test the trained model on the RNA?Puzzles dataset. The results show that，among 28 RNAs，the model correctly identify experimental structures out of decoys with a ratio of 71.4%，superior to our previous model. Furthermore，we analyze the underlying mechanism of the neural network,finding that the way it scores the structural elements is consistent with known physical?chemical principles.

The manufacturing industry produces massive multi?source heterogeneous data such as texts，images，audio and video in the process of design，production，sales and service. The major problem facing manufacturing companies is how to efficiently manage and use these data resources to create value for manufacturing reproduction. Traditional data storage and retrieval systems classify these data according to different forms or modalities and process them separately，resulting in a lack of correlation between cross?modal data (texts，images，audio and video data cannot be checked each other). It cannot support the problem of manufacturing business processes. In this paper，we design and implement an efficient and fast cross?modal retrieval system for multi?source heterogeneous data such as texts and pictures to realize efficient management and retrieval of multimodal data. Specifically，the system projects the these data into a unified high?dimensional semantic space for representation，generates semantic vectors，and stores the multi?source heterogeneous data in different modes according to different query requirements. Then，the system designs an efficient retrieval method of three?level structure + layered Unicom naive composition algorithm，and indexes the multimodal data according to the semantic vector to meet the semantic query needs of manufacturing users. We conduct experiments on the flickr30k dataset. Experimental results show that: (1) This system can support millions of data storage and retrieval. (2) With millions level data，the system retrieval rate is milliseconds. (3) The retrieval accuracy is higher than traditional vector retrieval methods.

Decision evolution set is a theory to deal with evolution problem of decision rules in time series. Decision evolution set transfers the focus from static decision information system to dynamic time series，which is a new decision research method to study the evolution regulations in decision information system following the time variation. In the decision evolution set theory system，the forecast rules are accompanied by the real rules，so that the forecast rules inevitably have an impact on the real rules. In order to explain the relationship between forecast rules and real rules，in this paper，the convolution method is used to construct the evolution mixed matrix of forecast rules and real rules on time series，and the matrix is used to forecast and analyze the decision information system.

Attribute reduction is one of the key issues of rough set theory. In order to satisfy different users' requirements for reduction，a dynamic attribute reduction method based on three?way decisions is proposed to adapt to the change of user preferences. First，incorporating multi?user preferences，a user preference matrix is defined to describe the preference of each attribute under multiple users. Then，combining attribute preference degree and cost in real problems，the user preference index is proposed to indicate the importance of attribute in the current user group，and used as heuristic information for attribute selection. Finally，the three?way decisions theory is used to divide the reduced set and non?reduced set into three parts so as to achieve the purpose of updating reduction by taking different strategy. Case analysis and experimental results verify the feasibility and effectiveness of the proposed method，and the obtained reduction can satisfy the requirement of multi?user well.

The South Pole?Aitken Basin (SPA) is one of the largest impact basins in the solar system，and the largest and oldest impact basin on the Moon. The SPA basin provides a critical example for investigating giant impact events on the Moon at its early evolution stage. Because the diameter and number of impact craters follow a power law，it is difficult to identify numerous small impact craters merely by human labor. In recent years，the improvement of computer computing power makes it possible to train complex convolutional neural networks. Automatic identification of craters can be realize by a trained neural network，simultaneously improving the efficiency and ensuring the accuracy of crater identification. In this study，we applied the You Only Look Once Version 5 (YOLO V5) target detection system based on the convolutional neural network algorithm to automatically identify small impact craters with a diameter range of 2~15 km in the SPA basin. For the model training，we used SELENE?LRO merged digital elevation model SLDEM2015 and the latest expert?labeled crater catalog. Compared with the expert?labeled crater catalog，the trained network gets a precision of 0.96，a recall of 0.95，and an F1?score of 0.95 on the test set. By verifying our identified craters inconsistent with the expert labeling，we find more than 10 impact craters mislabeled by the expert. This proves that the automatic crater identification method can be used to verify the reliability of expert labeling. Based on the automatically identified craters，we also determine the absolute model ages for four typical mid?sized craters，providing a useful application of the crater identification results. Our estimated absolute model ages are consistent with existing dating results. We expect that the automatic crater identification of this study will be extended to the identification of smaller craters，and be transferred to other geological units on the Moon and even other terrestrial planets and rocky satellites.

Volcanism is the manifestation of internal thermal activity，and the style and history of volcanism are important target to resolve the evolution of planetary surface enviroment and habitability. Similar with the other terrestrial bodies，the planet Mercury has been resurfaced by large?scale and long?term volcanic activities. However，characteristics of volcanism on Mercury，such as the compositions，volcanic landforms，eruption mechanisms，and active durations are obviously different from those of the Earth，Mars and Venus. Global effusive volcanism on Mercury occurred within the first billion years of its history and formed intercrater plains and smooth plains that cover a significant portion of the planet. Large?volume effusive volcanism was ceased at about 3.5 Ga due to interior cooling，thus the induced global contraction caused compressive stresses in the lithosphere，impeding the ascent of magma. Afterward，volcanism on Mercury has been shifted to small?scale explosive volcanism that is drived by volatile?rich magma along pre?existing crustal weakness. Explosive volcanism may continue to Mercury'?s recent history. The history of volcanism on Merury reflects special geodynamical process in the mantle，revealing the significant effect of basin?impacts on thermal evolution and providing insights into the origin and evolution of Mercury.

The existing hybrid structure learning algorithm is subject to the neighbor set of variables，which results in the hybrid structure learning algorithm in the constraint learning stage. If the neighbor set of variables does not contain nodes of real structure，the nodes will not be considered. To improve this problem，this paper explores the Bayesian network structure and the possibility of relation between node effect degree，designs a variable order adjustment method based on node effect degree，and applies the adjusted variable order to network structure learning. The adjusted variable order not only reduces the search space，but also improves the problem that the traditional constraint space relies too much on the neighbor set of variables，thus improving the learning quality of the network structure. Experimental results show that the proposed algorithm can effectively improve the accuracy of existing hybrid structure learning algorithms，and also verify the feasibility of exploring Bayesian network structure diagram from the point of node effect degree.

In current work，the inhibitory and killing effects of Antimycin on Triple Negative Breast Cancer cell MDA?MB?231 were studied. Three natural products with similar structure，Antimycin?1，?2 and ?3，had strong inhibitory effects on the growth of MDA?MB?231 cells. And the IC50 were 1.34±0.07，160±20 and 180±50 nmol·L^-1，respectively. The activity of Antimycin?1 was about one hundred times higher than that of Antimycin?2 and Antimycin?3. 10 nmol·L^-1 Antimycin?1 could effectively inhibit the proliferation of MDA?MB?231 cells and the inhibition rates were about 80% and 90% after 24 h and 48 h treatment. The morphology of MDA?MB?231 cells were observed after treated with different concentrations of Antimycin?1. 10 and 100 nmol·L^-1 of Antimycin?1 killed most MDA?MB?231 cells. 1000 nmol·L^-1 concentration of Antimycin?1 made the cells almost dissolve，leaving the nucleus and cytoplasmic debris. However，the morphology of MCF?10A cells and HCT116 cells were not significantly changed at the same concentrations of drugs. 20 and 50 nmol·L^-1 Antimycin?1 also significantly changed the nuclear morphology of MDA?MB?231 cells. The nucleus showed deformity，severe shrinkage and nuclear membrane damage. Treated with antimycin?1 at 5，10，100 nmol·L^-1 concentrations for 12，24 and 48 h，the numbers of apoptosis and necrosis of MDA?MB?231 cells increased in a time and dose?dependent manner. After treatment with Antimycin?1 at concentrations of 5，10 and 20 nmol·L^-1 for 6，12 and 24 h，no significant effects on cell cycle phase of MDA?MB?231 were observed. And the level of ROS (reactive oxygen species) in MDA?MB?231 cells decreased gradually with the prolongation of treatment time，after MDA?MB?231 cells were treated with 20 nmol·L^-1 Antimycin?1. These results showed that Antimycin?1 at nanomolar concentration could effectively inhibit and kill MDA?MB?231 cells.

Neural style transfer is a technique to transfer a style image to a content image using the principle of deep learning. Recently，the approaches to transfer image style have been successful. However，there is an important disadvantage when style transfer is applied to photorealism style: the image structure is prone to change and the image style is unnatural. In order to improve the quality of generated image in photorealism style transfer，we present a neural style transfer method which is based on convolutional neural networks. Our approach aggregates multiple features from the shallowest layers and deeper layers to represent style image features. With a combination of global style loss and local style loss as total style loss. The global style loss is Gram?based which uses Gram matrices to represent global style features and the local style loss is based on MRFs (Markov Random Fields). In order to restrain the change of image structure，we preserve edges by constraining the transformations locally affine in color space. And our approach presents a semantic segmentation module based on neural network. It automatically generates semantic segmentation of the input image to constrain the image distortions. Experimental results show that our approach improves generated images quality of photorealism style transfer in variety of scenarios.

Deep learning based on neural network has made remarkable progress in wide domains. However，the latest research shows that deep learning bears the risks of privacy leakage. The current defense mechanisms are limited by the assumption of the adversary's background knowledge，the high complexity and the non?universality. This paper aims to construct a novel defense mechanism for deep learning using differential privacy. At present，the widely used differential privacy algorithm is DPSGD (Stochastic Gradient Descent with Differential Privacy) in deep learning. However，parameter setting is difficult for DPSGD，and the measurement of privacy loss is also complex. We propose DPADAM (Adaptive Moment Estimation with Differential Privacy) as a new deep learning optimization algorithm with privacy protection，which combines Adam gradient optimization algorithm and differential privacy. Also，we introduce zCDP (Zero?Centralized Differential Privacy) as a measure of privacy loss，which is more flexible and accurate. Extensive evaluation results show that DPADAM can reduce dependence on parameter settings and improve the model's fitting effect.

The reduction of triadic concepts is an important problem in triadic concept analysis，because it can not only simplify the representation of triadic graphs，but also help to better understand the meaning of triadic concepts and extract valuable information from them. Based on the triadic factor analysis，reduction of triadic concepts is studied to keep all the triadic relationships in the triadic context. Firstly，a reduction of triadic concept is defined based on triadic factor analysis. The method is to find as few triadic concepts as possible under the condition of preserving the original triadic context. That is these selected triadic concepts can completely reflect all the triadic relations contained in the original triadic context. Secondly，the relationship between the triadic factorizations and the triadic concept consistent sets is discussed，and the necessary and sufficient conditions for consistent sets and reducts are given. Finally，the triadic concepts are classified into three categories by using triadic concept reduction: core （absolute necessary） concept，relative necessary concept and unnecessary concept. Moreover，the necessary and sufficient conditions for each class of triadic concepts are obtained. In addition，the detailed process of finding triadic concept reduct using triadic factorization and the definition of reduction is given by an example.

Rule acquisition is one of important research fileds of knowledge discovery. Multi?source data is an important data set，and obtaining rules in multi?source data from different perspectives can provide a more reliable basis for decision making. In this paper，based on multi?source decision tables，the definitions of two types of multi?source decision rules are presented from the perspectives of the data source and the conclusion of decision rule. Then，the support degree and the coverage degree of two types of rules are proposed. After that，the relationships between two types of multi?source decision rules are discussed. Finally，the approach to rule acquisition and rule simplification of the first and second types of multi?source decision rules are investigated by introducing the unisource decision table of multi?source decision table.

Raman spectroscopy is an in situ，fast，and nondestructive analytical technique，which is widely used for qualitative description，semi?quantitative and quantitative analysis of fluid inclusion components. The Raman OH stretching band of water (ν_s?H₂O ) shifts to higher wavenumber and becomes sharper and more symmetric with increasing NaCl concentration，which can be used to investigate the salinity of the aqueous phase inside the fluid inclusion. It should be noted that crystal orientation of the host mineral with birefringence also influences the ν_s?H₂O band shape，which limits the application of this method to the determination of fluid salinity in natural fluid inclusions. Calcite is one of the most common host minerals of fluid inclusions. However，experimental investigations on the effects of calcite crystal orientation on ν_s?H₂O band are lacking. In this paper，a series of fused silica capilary capsules (FSCCs) containing different concentrations of NaCl solutions were prepared. Raman spectra of these solutions were collected at room temprature and the quantitative relationship between NaCl concentration and Raman spectral parameter was established. Three calcite thin sections were prepared along the \left\{\mathrm{1011}\right\}，\left\{\mathrm{1121}\right\} and \left\{\mathrm{0001}\right\} orientations. The calcite thin section was placed on the FSCC containing water or NaCl solution. Then，the calcite thin section was rotated clockwise for 180°; Raman spectra were collected at intervals of 15°. The effect of calcite crystal orientation on the ν_s?H₂O band was investigated by comparing these spectra collected at different rotation degrees. This effect can be described as eihter the low wavenumber component or the high wavenumber component of ν_s?H₂O band was enhanced.Based on these experimental investigations，a feasible method was proposed for themeasurements of the salnity of fluid inclusions hosted in calcite. It is recommended that multiple Raman spectra should be collected at intervals of 15° when fluid inclusion wafer was rotated by 180°. Then，the average value of the spectral parameter was obtained and used to calculate salinity. At last，we measured the salinity of fluid inclusions in the late hydrothermal stage calcite in vein?type tungsten deposit at Yaogangxian by applying the microthermometric and in situ Raman spectroscopyic methods. The results show that the salinity of natural calcite?hosted fluid inclusion can be measured using in situ Raman spectroscopic method. The salinity deviation obtained by these two methods was reported to be within ±5%.

The degradation of organic matter in microbial carbonates during early diagenesis not only can produce acidic fluids to dissolve carbonate grains，which may amplify the channels for later dissolution and be beneficial to the development of reservoirs，but also can result in alkaline pore water that precipitates carbonate cements which may hinder the development of reservoir. Therefore，the degradation of organic matter during syngenetic?early diagenesis stage has important impacts on the development of reservoir in microbial carbonates. However，the characteristics and recognization about degradation of organic matter in microbial carbonates have not yet been studied. In this study，we conducted petrological analyses and in situ geochemical for microbial carbonates from the Member IV of Dengying Formation，Northern Sichuan Basin to characterize organic matter degradation in microbial carbonates. Results show that when the organic matter in the microbial carbonates is degraded by aerobic respiration，both the microspar and sparite components have negative Ce anomalies，whereas the positive Ce anomaly of sparite component indicates that the degradation of organic matter is under an anaerobic environment. If organic matters in microbial carbonate are oxygenated by nitrate reduction，the Cr concentration of microspar component is lower than that in sparite component. If the organic matter is degraded by Fe?Mn oxides reduction，the Fe concentration in microspar component is higher than that of sparite component. With the reduction state of diagenetic environment further strengthening，the organic matter may be further degraded by sulfate reduction，which leads to the higher Cu and Mo concentrations in the sparite component compared with the microspar component in microbial carbonates. As a consequence，the variation of elements Ce，Cr，Fe，Mo，Cu in components of microbial carbonates could reflect the diagenetic environment，and also can trace the process of the organic matter degradation effectively.

At present,Brain?Computer Interface (BCI) based on Motor Imagery (MI) provides relatively few instructions. To add new control parameters,this paper studies the single?trial recognition of actual and imagined forces of hand clenching based on Electroencephalogram (EEG). Twenty subjects recruited to participate in the experiment. They were instructed to perform three different actual/imagined hand?clenching force tasks (4 kg,10 kg,16 kg) with their right hand. The EEG data of nine channels over the primary motor area and the supplementary motor area during the task were analyzed,and the Common Spatial Pattern (CSP) was used to extract features of actual/imagined force of hand clenching,which were single?trial recognized by two classifiers: Extreme Learning Machines (ELM) and Support Vector Machines (SVM). The average ELM single?trial recognition accuracy for three different actual/imagined hand?clenching force tasks was 82.3%±2.1% and 80%±1%,respectively. The average SVM single?trial recognition accuracy for three different actual/imagined hand?clenching force tasks was 86.3%±5.5% and 83.7%±3.8%,respectively. The results show that the ELM and SVM can effectively identify three different actual/imagined hand?clenching force tasks,and the classification result of SVM is better. This study is expected to provide a new idea for adding new control parameters to MI?BCI.

With the fast development of Internet applications，it's common for someone to have accounts on different social network platforms. How to find out which account on multiple social networks are of the same person is an important issue for many applications today，which is also known as the user alignment problem. There are two major challenges when it comes to user alignment. First，it's extremely expensive to collect manually aligned user pairs as training data，but traditional supervised methods often need a large amount of labeled data to achieve better results. Second，users on different networks often have different structures and attributes，which further increase the difficulty of user alignment. We propose an unsupervised user alignment method SPUAL (Soft Principle for User Alignment)，design a novel soft alignment principle based on user attributes and structure，and then infer whether the user alignment is correct or not by calculating whether the users meet to the principles by unsupervised method. Experiments on several common datasets show that the performance of our method is much better than the most advanced unsupervised methods.

Montmorillonite is the most commonly distributed clay mineral in earth surface system. Exchangeability of interlayer cation is one of its characteristic properties，which makes it an important natural materials for various applications. Different cations (K⁺，Mg²⁺，Ca²⁺，Ba²⁺) exchange from montmorillonite interparticle pore fluid into clay interlayer space has been studied at atomistic level by using classical molecular dynamics simulation. The final cation exchange capacities follow the order of Ba²⁺>Ca²⁺>K⁺>Mg²⁺，as opposed to the hydration ability of cations. In the montmorillonite interlayer，Mg²⁺ is absolutely hydrated by waters and far away from the montmorillonite surface. Parts of Ca²⁺ and Ba²⁺ ions may closely hover above the tetrahedral substitution position of the clay，however，most K⁺ ions are bound by the six?membered ring of the clay surface. The migration of cation from fluid to clay interlayer space corresponds a process with decrease in free energy. The mobility of interlayer cations is much lower than that in fluid，especially the self?diffusion of Ba²⁺ is the lowest. The disclosed dynamic process of cation exchange of montmorillonite at atomistic level will enhance the understanding of clay?fluid interactions.

Named entity recognition of electronic medical record is a key basic task in the field of medical artificial intelligence and medical information service. In order to fully explore the semantic knowledge of medical entity in electronic medical records to improve the effect of Chinese medical entity recognition，BiLSTM(Bi?Long Short Term Memory)?withfea model incorporating external semantic features is proposed. The model first generates a character?level vector with semantic features using the large?scale unlabeled medical texts based on word2vec. Second，a medical entity and feature database for entity recognition is constructed according to the integration of the existing medical resources and the analysis of different types of entity boundary characteristics. Then，the database is converted into vectors to join with character?level vectors to fully exploit the sequence information. Finally，the experimental result is integrated with the output of the CRF (Conditional Random Fields) by the improved Voting algorithm to correct the label offset. Experiments show that the F?measure of the improved model reaches 94.06%，which is 1.55% higher than CRF. In addition，the parameters of the optimal effect of the improved model are given.

The complexity of urban traffic road network and the China's car parc are growing at a high speed. Traffic jam now is a common problem in most citys. And the quality of people's driving trips is being seriously affected. How to plan a real?time and optimal driving route in a complex and variable traffic network has become a real concern for people when they travel. In most of the current research on the optimal route selection of traffic，only the static traffic road network scenario is considered，and the cost of driving through the signal intersections is neglected，which resulting in a large error between the calculation result and the actual driving cost. To solve this problem，a more accurate multi?factor traffic network model with road and signal intersection based on Petri nets is built，and a traffic optimal route selection algorithm based on improved elitist ant system is proposed. Improvements on the classical ant colony optimization are provided in two aspects. Firstly，the main road guiding and driving direction guiding are added in the initialization of the pheromone concentration to speed up the initial search; Secondly，a dual elitist ant strategy is utilized that the pheromone concentration on the two optimal paths is globally updated in mutually constraints way. This strategy can accelerate the convergence of the algorithm while avoiding the local optimal solution and two optimal paths can be searched to for selection. The simulation results show that the probability of finding the optimal path is increased to 100% while the convergence rate is guaranteed with the proposed algorithm. At the same time，the convergence rate of this algorithm is several times of others’ under the premise that the probability of finding the optimal solution is not less than 90%.