384 results about "Objective model" patented technology

A method and system are provided for targeting ads by effectively combining behavioral targeting and social networking. In one example, the method includes receiving a behavioral targeting model to predict a propensity of each consumer in a network to select (e.g., click) an ad of a particular category based on a behavior of each consumer, training a social network model to predict a propensity of a particular consumer to select an ad of the particular category based on features derived from a social network of the particular consumer, and training an ensemble classifier to decide when to trust the behavioral targeting model and when to defer to the social model for predicting a propensity of the particular consumer to select an ad of the particular category.

A method of determining an overlay error. Measuring an overlay target having process-induced asymmetry. Constructing a model of the target. Modifying the model, e.g., by moving one of the structures to compensate for the asymmetry. Calculating an asymmetry-induced overlay error using the modified model. Determining an overlay error in a production target by subtracting the asymmetry-induced overlay error from a measured overlay error. In one example, the model is modified by varying asymmetry p(n′), p(n″) and the calculating an asymmetry-induced overlay error is repeated for a plurality of scatterometer measurement recipes and the step of determining an overlay error in a production target uses the calculated asymmetry-induced overlay errors to select an optimum scatterometer measurement recipe used to measure the production target.

The invention provides a deep learning adversarial attack defense method based on a generative adversarial network. The method comprises the following steps: step 1), based on the high performance ofthe generative adversarial network in learning sample distribution, designing a method for generating an adversarial example through the generative adversarial network, and after adding a target modelnetwork set TMi, enabling the sample generation based on a G network to become a multi-objective optimization problem; and the training for an AG-GAN model is mainly for the parameter training of thegenerative network G and a discrimination network D, and is divided into three modules; and step 2), using the adversarial example generated by the AG-GAN to train an attacked deep learning model, soas to improve the capability of the deep learning model of defending different types of adversarial examples. The deep learning adversarial attack defense method based on the generative adversarial network provided by the invention effectively improves the security.

A system and method for developing and enabling model-driven extensible Markup Language (XML) transformation to XML Metadata Interchange (XMI) format incorporate a strong built-in validation capability. A platform independent framework applies multiple passes of transformation, where each pass performs specific operations on internal models. Different source models are then merged into a target model.

The invention discloses a model parameter training method and device based on federal learning. The method comprises the steps of using a first terminal to receive a first encryption mapping model sent by a second terminal; predicting the missing feature of the first sample data according to the first encryption mapping model to obtain the first encryption complement sample data; training a federal learning model according to the current encryption model parameters, the first sample data and the first encryption completion sample data, and obtaining a first secret sharing loss value and a first secret sharing gradient value; and if it is detected that the federal learning model is in a convergence state, obtaining a target model parameter according to the updated first secret sharing model parameter corresponding to the first secret sharing gradient value and a second secret sharing model parameter sent by the second terminal. According to the method, by adopting a secret sharing mode, the training process of the federated learning model does not need the assistance of a second collaborator, and the user experience is improved.

A geometric model comparator is provided, which includes processing circuitry, memory, and comparison circuitry. The processing circuitry is configured to generate a target model from a source model. The memory is configured to store the source model and the target model. The comparison circuitry is configured to identify selected points from the source model, create corresponding selected points in a target model, and compare the selected points from the source model with the selected points from the target model to identify one or more selected points from the target model that fall outside of a predetermined tolerance range with the respective one or more points from the source model. A method is also provided.

The invention discloses an eye-tracking method and device and belongs to the technical field of image processing. The eye-tracking method comprises the following steps: determining an observation area where the iris center of an iris image to be tested is according to a target model which is obtained according to target parameters and an extreme learning machine neural network; adopting the observation area to correct a predication area to obtain a target area, wherein the prediction area is the area, where the iris center of the iris image to be tested is located, determined by the Kalman filtering method; determining a screen falling position watched by an eye according to the target area. The eye-tracking method and device solve the problems that the eye watching accuracy and speed of the screen falling point are relatively low, achieve the effect of improving the eye watching accuracy and speed of the screen falling point, and are used for eye tracking.

The invention discloses a model parameter training method and device, a server and a storage medium, which belongs to the technical field of information. The method comprises the steps that an initialparameter value and a sample set of a model parameter of a target model are acquired; the first gradient of the model parameter is calculated according to the initial parameter value and the sample set; iterative quantization processing is carried out on the first gradient to acquire a quantized second gradient, wherein the iterative quantization processing is quantization processing carried outbased on an error cumulative value corresponding to the t-1-th iteration round in the t-th iteration round, and the error cumulative value is a quantization error cumulative value calculated based ona preset time attenuation coefficient; and the quantized second gradient is transmitted to a primary computing node, wherein the quantized second gradient is used to instruct the primary computing node to update the initial parameter value according to the quantized second gradient to acquire an updated parameter value. According to the embodiment of the invention, a quantization error correctionmethod is used to quantize and compress the first gradient of the model parameter, which reduces the communication cost and network overhead of gradient transmission.

The invention discloses a video target tracking method based on CAMSHIFT and Kalman filtering. The method takes a search window as a parameter and comprises the following steps of: predicting a search window Win by a Kalman filter; calling a CAMSHIFT algorithm by taking Win as a parameter to search for the target, and returning to the window targetWin containing candidate models; calculating similarity between the targetWin and the target model; if the similarity is greater than the set threshold, finding the target and returning to the target window; otherwise, taking the target window of the previous frame and the distance predicted by the Kalman filter as a window expanding parameter instead of enlarge, and calling an adaptive iterative search algorithm. By introducing the similarity judgment and the re-positioning technology, the method effectively eliminates the background interference and improves the tracking precision.

The invention discloses an adversarial-learning-based multi-source-domain adaptive migration method and system. The method comprises: step one, pre training is carried out by using all-source-domain data and a representation network and a classifier of a target model are initialized; step two, multi-path adversarial adversarial processing is carried out on multi-source-domain data and target-domain data and a representation network and a multi-path discriminator of the target model are updated; step three, adversarial scores between the source domains and the target domain are calculated; stepfour, target domain classification is carried out based on the classifiers and the adversarial scores of all source domains; step five, a target domain pseudo sample with a high confidence coefficient is selected for fine tuning of the representation network and the classifier of the target model; and step six, the steps from the step two to the step five are carried out again until model convergence is realized or a maximum iteration number of times is reached, and then training is stopped. According to the invention, reliance on the hypothesis of consistency of the single-source-domain tagset and the target domain is eliminated; and a negative migration phenomenon existing in the multi-source domain adaptation process is avoided effectively.

The invention discloses a video target tracking method capable of integrating ASIFT features and particle filter, and belongs to the technical field of video information processing and mode recognition. The video target tracking method comprises the following steps: the adjacent frame difference method is utilized to obtain moving objects in a video sequence; according to the area corresponding to an acquired complete target, a tracking target model is established; ASIFT feature vectors of the target model are established; the particle filter technology is adopted to predict a candidate area target; ASIFT feature vectors of a candidate target model are established; the feature vectors of the tracking targets are matched with the feature vectors of the candidate target; the RANSAC algorithm is adopted to reject wrong matching; the target model is renewed, so as to realize target tracking. The video target tracking method provided by the invention can accurately and quickly track the targets under the condition that brightness changes and is shielded. Therefore, the multi-target tracking method in the video surveillance has relatively good real-time performance and robustness.

A system maps data within a data source to a target data model, and comprises a computer system including at least one processor. The system determines an identifier for each data object of the data source based on the data within that data object, wherein the identifier indicates for that data object a corresponding concept within a domain ontological representation of a data model of the data source. The determined identifiers for the data objects of the data source are compared to the target data model to determine mappings between the data objects of the data source and the target data model. Data objects from the data source are extracted for the target data model in accordance with the mappings. Present invention embodiments further include a method and computer program product for mapping data within a data source to a target data model.

A method of determining a rigid motion between a master solid model and an approximated target model (or, more generally, between any two models having different types) includes identifying, within each model, geometrical entities having a unique characteristic, and then determining the best match between the identified geometric entities. The system provides, in machine-readable form, a master model comprising a precise definition of a three-dimensional solid and a target model comprising a simplified definition of the three-dimensional solid. Then it identifies a first set of geometric entities (e.g., planar faces) within the master model that have a unique characteristic (e.g., planar area), and identifies a second set of geometric entities in the target model that have the unique characteristic. The system then determines a best match between a member of the first set of geometric entities and a member of the second set of geometric entities using, for example, a Hungarian matching algorithm. Linear edges of matched faces are compared to determine the appropriate rigid motion.

The invention is applicable to the electronic field and provides a target identification method and a device. The method comprises the following steps of: judging whether a starting action exists in ROI (region-of-interest) of a back depth map according to variation of depth value of a front depth map and the back depth map which are adjacent in a depth frame sequence; detecting regions which have the same colour frames according to a preset limb target model, and judging regions which accord with the limb target model to be limb target regions; storing characteristic set parameters of the limb target regions; tracking regions of the previous colour frame which are judged to be the limb target regions in a depth frame, and detecting the regions which are the same in the colour frame corresponding to the preset colour frame by utilizing the preset limb target model and characteristic set parameters of the previous limb target region which are stored, so as to obtain the limb target regions; acquiring coordinates of each limb target region, and identifying a target action according to the acquired coordinate values. In the target identification method and device provided by the embodiment of the invention, a depth image sequence and a colour image sequence are used for detecting the limb target regions, thus detection accuracy is effectively improved.

The invention relates to an image recognition attack method based on algorithm confrontational attack. The method includes inputting the original image needing to be identified and attacked into the adversarial generation network to obtain a resistance image, carrying out image identification and classification on the original image and the resistance image simultaneously, if the classification isthe same, indicating that the attack is unsuccessful, collecting data and updating the adversarial generation network, otherwise, indicating that the attack is successful. According to the method, anexisting image recognition algorithm can be attacked, the algorithm cannot carry out normal image recognition by generating a resistance sample, and therefore functional application in the fields offace recognition, image detection, automatic driving and the like is influenced, and the applicability is wide; once the training of the adversarial generation network is completed, the generated adversarial samples do not need to depend on the contact of a target model and a large number of numerical operations, and the characteristics of high efficiency and migration are achieved; research on the adversarial attack of machine learning is beneficial to further optimization of a machine learning algorithm and a data processing means, and the safety of the machine learning algorithm and the application thereof is improved.

The invention relates to an online training method, device and system and a computer readable storage medium, and belongs to the field of deep learning. The method comprises the following steps: dividing all model parameters of a deep neural network model into a plurality of parts, and when real-time training sample data is obtained, training the deep neural network model by adopting model parameters corresponding to the real-time training data to obtain gradient information of each model parameter; Acquiring at least one target model parameter based on the gradient information of each model parameter and the model parameters corresponding to the real-time training data; And based on the at least one target model parameter and the gradient information of each model parameter, updating thedeep neural network model to achieve the purpose of on-line model training.

The embodiment of the invention discloses a recognition method for the operating state of industrial equipment and a server. The method includes the steps that the server acquires a first training set; if the server inputs the first training set into a selected first target model, the server uses a first target algorithm to train the first target model , and a second target model is obtained; if the server determines that the accuracy of the second target model does not reach a preset accuracy threshold, the server adjusts hyper-parameters of the second target model according to a preset rule;if the server inputs the first training set into the second target model with the hyper-parameters adjusted, the server uses the first target algorithm to train the second target model with the parameters adjusted, and a third target model is obtained; the server judges whether or not the accuracy of the third target model reaches the accuracy threshold, wherein the accuracy of the third target model is obtained by the server according to a first test set; if yes, the server uses the third target model to predict the operating state of the industrial equipment.

A method, system and apparatus for the model-to-model transformation by element-kind of a source model are proposed. An element-kind model-to-model transformation can include one or more transform elements defining a traversal of a source model, an element-kind mapping and the conversion from source models to target models. The element-kind mapping can include a set of associations between element-kinds for the source model and corresponding transformation rules to produce target objects in a target model.

The present invention, belonging to the technical field of computer graphics and computer assistant graphic design. The method according to the present invention comprises: firstly inputting a three-dimensional garment model set, through shape and style analysis, segmenting the three-dimensional garment parts having the same style structure to achieve semantic segmentation; clustering the three-dimensional garment part models upon the segmentation into four major categories, to form a three-dimensional garment part library; defining a style description measuring recombination of garment parts by using the area of the garment part model and the boarder circumference ratio as major geometric shape features; performing global optimization for the source models of the three-dimensional part model clustering according to the defined style descriptor and target model; and finally outputting a new three-dimensional garment by means of natural splicing. The method according to the present invention avoids the complicated process of garment modeling and improves the efficiency of three-dimensional garment modeling, effectively meets the needs of the present large-scale three-dimensional garment quantity, and has the advantages of practicality and effectiveness.

A method for searching a monitor video target comprises the steps as follows: step 1, obtaining a searched target image and extracting feature of the searched target image so as to establish a target model; step 2, obtaining an input video and establishing a background model according to the input video so as to extract a moving foreground in each video frame and detecting to obtain a candidate target area via a connected domain, calculating a possible object target frame position by using an edge outline estimation method in the candidate target area; step 3, using the feature similarity of an object sub-frame of the candidate target area, which has maximum comprehensive similarity with a search target as similarity corresponding to the candidate target area; storing relative information of the candidate target area in a database if the comprehensive similarity of the candidate target area and the search object is more than a threshold value; step 4, classifying and selecting record in the database so as to reduce the emerging times of a repeated object in a continuous time period in the selecting result; step 5, inquiring the database and ranking search results according to a preset condition.

The invention discloses a video scene recognition method and device, a storage medium and an electronic device. The video scene recognition method comprises the steps: obtaining a target video to be subjected to scene recognition; performing frame extraction sampling on the video sequence of the target video to obtain a sub-video sequence, and performing frame extraction sampling on the audio sequence of the target video to obtain a sub-audio sequence; extracting a target video feature from the sub-video sequence, and extracting a target audio feature from the sub-audio sequence; processing the target video features through the first target model to obtain a first classification result, and processing the target audio features through the second target model to obtain a second classification result; and determining a target scene of the target video based on the first classification result and the second classification result. According to the invention, the technical problem of low recognition accuracy of the video scene in the prior art is solved.

The invention relates to a target tracking method and a system based on an on-line initialization gradient enhancement regression tree. In the system consisting of a video input end, a tracking target output end and an on-line training classifier, the method comprises the steps of 1) selecting a tracking target from a video series and extracting positive and negative samples of a Haar-like feature, 2) randomly establishing the on-line classifier according to the positive and negative samples to obtain a training residual error, 3) conducting training amendment by taking the training residual error as a training sample of the on-line classifier and establishing a target model, and 4) acquiring an image confidence map from a next frame of video image, determining a maximum position of a confidence value in a target window, and accomplishing tracking. According to the method and the system, the tree can be converged to an optimum point quickly to ensure that the random forest detection optimization is accomplished, the classifier is updated through on-line study, and the problems of appearance variation, rapid movement and shielding of a target are solved well.

The invention discloses an image recognition method based on a depth neural network model. The target image is inputted into the target model which is obtained by pruning the depth neural network model with the channel representation ability. Using the auxiliary classifier in the target model to classify the target image, the recognition results are obtained. Because the target model of target image recognition is a pruned model based on the channel representation ability, the computational cost of target image recognition can be greatly reduced. The invention also discloses an image recognition device, a device and a readable storage medium based on a depth neural network model, which have corresponding technical effects.

The invention discloses a composition modeling method for a combustion system based on numerical simulation and test operation data, which belongs to the technical field of composition modeling during combustion of a utility boiler combustion system. The method comprises the following steps: a numerical simulation model of a target coal-powder boiler is built by adopting the three dimensional steady-state condition; according to a target model built by the LS-SVM, firstly applicable input variables are determined, and the value range of each input variable is determined; by utilizing the given sampled data group count of the orthogonal method and the value of each input variable, a variable list and testing data are obtained and the mechanism LS-SVM model is established; the LS-SVM model is updated according to the real-time data of the power plant. The method overcomes the defect that the data is single during the modeling of simple test data, updates the model by utilizing the data updated by the power plant in real time, effectively improves the precision of the LS-SVM model, and lays a foundation for optimization of the boiler combustion.

The invention discloses a target tracking method based on an LSTM neural network, and belongs to the technical field of target tracking. According to the method, the LSTM is used for tracking the complex and non-linear motion target, and the problems that target tracking is difficult, a target model is difficult to establish and the tracking precision is low are solved; the method comprises the following steps: firstly, acquiring latitude and longitude information and speed information of a target, and processing acquired data; designing an LSTM neural network structure for single target tracking; and finally, adjusting LSTM neural network parameters to realize target tracking. According to the method, the nonlinear filtering process is effectively simplified, and a complex nonlinear target can be effectively tracked; the establishment of a target motion model and the utilization of a traditional filtering algorithm are not needed; estimating the target motion state of the next momentby using historical target motion information; adjusting internal parameters of the neural network by using a back propagation algorithm; the learning rate attenuation method reduces the calculation amount and improves the precision.

The invention relates to an image detection model training method and device and a storage medium, and the method comprises the steps: obtaining a sample image set for training a target model; determining a category regression loss function of the target model for a target sample image according to the number of samples corresponding to different sample categories contained in the sample image setand the prediction probability of the target model for the currently input target sample image; and for each sample image in the sample image set, adjusting a category regression loss function of thetarget model, and training model parameters in the target model through the sample images. The technical problems that under the condition that samples are unbalanced, training tasks are difficult toconverge, and the accuracy and recall rate of a model obtained through training are not high are solved. The method has the advantages that the convergence speed of the sample categories with the small sample number is increased, and the accuracy and recall rate of the model obtained through training are increased.

The invention discloses an opportunistic network link prediction method and device, and a readable storage medium. The method comprises: slicing sample data to obtain a plurality of sub-sample data, and dividing a connection number and a connection time of a node pair in each sub-sample data into a corresponding network snapshot; converting the connection number and the connection time of the nodepair in each network snapshot into a connection weight, and mapping an identifier of the node pair and the node pair connection weight in each network snapshot into a time series vector sequence; acquiring node attribute information in each network snapshot to construct an attribute vector, and constructing an attribute vector sequence; constructing a Bayesian deep learning model, training and testing the model by using the time series vector sequence and the attribute vector sequence to obtain a target model, and predicting the opportunistic network link by using the target model. The methodand the device can accurately grasp the evolution rule of opportunistic network node pairs in the time domain and accurately predict the opportunistic network link.

An embodiment of the present invention provides a data query method, a device and electronic equipment. The method comprises: querying requests based on DSL data, getting the identity of the target model and the properties to be queried, identifying bya target model, obtaining the target model from a pre-stored model, according to the attributes to be queried and the target model, getting a querymodel, further, based on the preset mapping table between model attributes and fields, obtaining a mapping relationship between the attribute to be queried and the corresponding field to be queried, according to the syntax of the SQL statement, converting the mapping relationship between the query model and the attributes to be queried and the corresponding fields to be queried into SQL statements, the automatic generation of SQL statements is realized, which effectively reduces the workload of developers and improves the work efficiency. Moreover, since developers do not need to directly write SQL statements into the system, the risks of SQL statements being injected with viruses and the like are avoided, and the security of the system is improved.

The invention discloses a video target multi-target tracking method and device and a storage medium. The video target multi-target tracking method comprises the following steps: obtaining a first image feature of a target object and a second image feature of an observation object; performing feature similarity calculation of N feature categories on the first image feature and the second image feature to obtain N groups of feature similarity results; screening the N groups of feature similarity results based on the feature similarity of the rough set, and fusing the screened results to obtain afeature fusion result; based on maximum entropy intuitionistic fuzzy clustering, performing association cost matrix calculation according to the feature fusion result to obtain an association cost matrix calculation result; judging whether the target object is associated with the observation object or not according to the association cost matrix calculation result, and if yes, updating a target model; and if not, performing target trajectory management on the target object.

The invention provides an object tracking method and an object tracking system based on local classification. The object tracking method comprises the steps of acquiring a positive sample and a negative sample based on a first frame, and dividing into a training sample set and a verification sample set; performing local block sampling on the training sample set and training a local classifier of the object and acquiring an objective model; utilizing a candidate in subsequent frame pictures based on a particle filtering frame, performing estimation on each candidate by means of a local block and a corresponding local classifier, and updating the object position to the position of the candidate with maximal confidence. Therefore the invention provides the object tracking method and the object tracking system which overcome a drift tracking problem in a blocked condition, wherein the drift tracking problem cannot be settled based on global object tracking. Furthermore the invention provides a sample updating mode based on double-threshold restriction. Furthermore the weight of the local classifier is updated based on time domain stability of the object. Partial shielding can be effectively determined and suppressed, and furthermore randomness and contingency are prevented.