803results about How to "Improve generalization" patented technology

Site optimizer is a tool that enables A/B testing of merchandising offers for your online store to help you determine which offers work best on your storefront. A/B testing compares multiple offers simultaneously to randomly selected groups of shoppers. The A/B testing approach provides a unique and effective way to optimize online store performance because it is metrics-driven, objective, and more efficient than other site optimization techniques.

The invention belongs to the field of network resource management, relates to the cooperative filtration technique of network resources and discloses a network resource personalized recommended method based on an ultrafast neural network. The network resource personalized recommended method based on ultrafast neural network comprises the following steps: firstly, data preprocessing, reading information from the journal files of a system user and generating a global user interested matrix, exchanging the global user interested matrix into a single user interested matrix of a current user, and then transforming and reducing dimensionality to mark out a training set A1 and a predicting set A2, secondly, model training, building an interest predicting model with single hidden layer neural network SLFN s structure for a target user, adopting ultrafast learning machine technique to carry out training on the training set A1 and getting various connection power values and hidden layer threshold values of the neural network model with single hidden layer, thirdly, prediction recommending, utilizing the obtained predicting model to calculate the scoring values of every resource in the predicting set A2 given by the target user and then recommending the first several resources with highest predicting scores to the target user.

The invention discloses a hybrid machine learning credit scoring model building method. The method comprises the steps of 1, determining client risk classification standards based on a loan client historical data set; 2, based on the loan client historical data set, obtaining a loan client data feature set through feature extraction; 3, selecting at least two model algorithms from an alternative model library, building corresponding models based on the selected algorithms, performing model performance check on the built models by adopting a K-fold cross check method, performing standard check on the models about to pass the model performance check based on model check standards, obtaining evaluation index statistical quantity values, and selecting a model type used by final modeling according to the evaluation index statistical quantity values returned by the standard check of the models; and 4, based on the algorithm corresponding to the selected model type, building a credit scoring model. The technical effect of efficiently and accurately finishing user credit evaluation through the built hybrid machine learning credit scoring model is achieved.

The invention discloses a motion recognition method based on a three-dimensional convolution depth neural network and depth video. In the invention, depth video is used as the object of study, a three-dimensional convolution depth neural network is constructed to automatically learn temporal and spatial characteristics of human body behaviors, and a Softmax classifier is used for the classification and recognition of human body behaviors. The proposed method by the invention can effectively extract the potential characteristics of human body behaviors, and can not only obtain good recognition results on an MSR-Action3D data set, but also obtain good recognition results on a UTKinect-Action3D data set.

The invention relates to a conditional generative adversarial network-based monocular image depth estimation method. The method comprises the following steps of: (1) preprocessing a data set, (2) constructing a generator in a generative adversarial network: constructing the generative adversarial network formed by a convolution layer and deconvolution layer by utilizing an encoder-decoder structure, constructing a jump connection structure on the basis, mapping each layer of output feature map of an encoder into an input of a symmetric decoder, connecting the output feature maps through a channel dimensionality so as to increase detail information of the decoder, and sharing information of low layers between an input layer and an output layer to ensure that high-layer output images has low-level detail features and enhance the quality of the generated depth images; (3) constructing a judger in the generative adversarial network; (4) constructing a loss function of the generative adversarial network; and (5) training and testing the constructed generative adversarial network.

A method is described for improving the prediction accuracy and generalization performance of artificial neural network models in presence of input-output example data containing instrumental noise and/or measurement errors, the presence of noise and/or errors in the input-output example data used for training the network models create difficulties in learning accurately the nonlinear relationships existing between the inputs and the outputs, to effectively learn the noisy relationships, the methodology envisages creation of a large-sized noise-superimposed sample input-output dataset using computer simulations, here, a specific amount of Gaussian noise is added to each input/output variable in the example set and the enlarged sample data set created thereby is used as the training set for constructing the artificial neural network model, the amount of noise to be added is specific to an input/output variable and its optimal value is determined using a stochastic search and optimization technique, namely, genetic algorithms, the network trained on the noise-superimposed enlarged training set shows significant improvements in its prediction accuracy and generalization performance, the invented methodology is illustrated by its successful application to the example data comprising instrumental errors and/or measurement noise from an industrial polymerization reactor and a continuous stirred tank reactor (CSTR).

The invention discloses a head posture estimation method based on depth information and a color image. The method comprises the steps of firstly utilizing a histogram projection method to quickly position a head position in a depth information grey-scale map; then utilizing skin color detection algorithm to eliminate hair noise and clothes noise except the skin in the head area in a color image; utilizing a camera imaging principle to normalize the head area resolution, and homogenizing the depth information grey-scale map; and finally extracting a section regional geometric vector set from the obtained normalized grey-scale map to describe the variation rule of the head posture. According to the features of the depth information grey-scale map, the geometric feature of a face and particularly the symmetrical characteristics of the face about a nose tip, a novel head posture description characteristic code is designed, the head posture is estimated by utilizing a random forest regression machine, and the method has the advantages of good real-time property, high precision, high robustness and high reliability and can be widely applied to the application fields such as human face recognition and human-machine interaction.

A system discovers the topology of devices in a network, such as all of the layer-2 and layer-3 devices, including routers and switches. The devices are mapped, as well as the links that join them. The devices are polled using SNMP requests. Route tables are derived from each router to determine addresses of the devices. A spanning tree process is used to discover layer-2 devices. Connections between layers are then discovered. Where information is incomplete, an undiscovered router cloud is recognized.

The invention discloses a rolling bearing fault diagnosis method based on improved variational model decomposition and an extreme learning machine. The method comprises: vibration signals of a rollingbearing under different types of faults are collected, the vibration signals are filtered by means of maximum correlation kurtosis deconvolution, parameter optimization is carried out on the maximumcorrelation kurtosis deconvolution method by using a particle swarm algorithm, and an enveloped energy entropy after signal deconvolution is used as a fitness function; the mode number of variationalmodel decomposition is improved by an energy threshold and improved variational model decomposition of the filtered vibration signals is realized to obtain mode matrixes of the corresponding vibrationsignals; singular value decomposition is carried out on the mode matrixes to obtain a singular value vector and a rolling bearing fault feature set is constructed; and the fault feature set is trained by using an extreme learning machine and a rolling bearing fault diagnosis model is established. Therefore, stable feature extraction of the complex vibration signal of the rolling bearing is realized, so that the diagnostic accuracy is improved.

The invention discloses an intelligent detection and quantitative recognition method for the defect of concrete. According to the method, a concrete test piece is subjected to impact echo signal sample acquisition, signal noise reduction treatment and characteristic value extraction so as to construct a recognition model for analysis components including feature extraction, defect inspection, defect diagnosis and defect quantification and positioning; and the model is used for detecting and recognizing to-be-detected concrete. The intelligent detection and quantitative recognition method provided by the invention is directed at disadvantages of conventional detection technology for concrete defects and based on theoretical analysis, value simulation and model testing, employs advanced signal processing and artificial intelligence technology and fully digs out characteristic information of a testing signal, thereby establishing the model for intelligent rapid detection and classified recognition based on wavelet analysis and an extreme learning machine; and the model has good classified recognition performance, realizes intelligent rapid quantitative recognition and evaluation of the variety, properties and scope of the defect of concrete and further improves the innovation and application level of non-destructive testing technology for the defect of concrete.

The invention discloses an on-line fault diagnosis method of weighted extreme learning machine sewage treatment on the basis of a kernel function. The on-line fault diagnosis method comprises the following steps: 1) rejecting data with an incomplete attribute in sewage data, and then, carrying out data normalization processing to determine a historical data set and an update test set; 2) selecting a kernel function and a weighting scheme, and then, determining model parameters according to an optimal model; 3) according to the selected weighting scheme, endowing a weight for each sample of the historical data set; 4) training the model, and calculating a kernel matrix according to the kernel function; 5) adding a new sample into the model from a new test set for testing, and updating the historical data set; and 6) returning to 3), training the model again, and continuously repeating the above process until the on-line data test is finished so as to realize the identification of the on-line operation state of the sewage treatment process. The method has the advantages of short update time and high classification accuracy rate and has an important meaning for diagnosing operation faults in real time, guaranteeing the safe operation of sewage treatment works and improving the operation efficiency of the sewage treatment works.

The invention relates to the technical field of computer vision, and provides a pulmonary nodule image classification method when uncertain data is contained in a data set. The method comprises the following steps: firstly, collecting a pulmonary nodule CT image set, determining the category of the image through a majority voting principle by utilizing an expert voting method, and preprocessing toobtain a pulmonary nodule CT image data set; then, based on a knowledge distillation method, constructing a pulmonary nodule image classification model comprising a teacher model and a student model;next, obtaining a determined tag data set, training a teacher model on the determined tag data set, and calculating a soft tag on the pulmonary nodule CT image data set; then, training a student model on the data set combining the hard label and the soft label; and finally, inputting the preprocessed CT image to be classified into the trained lung nodule image classification model to obtain the category of the lung nodule image classification model. According to the method, the uncertain label data in the data set can be effectively utilized, the accuracy and efficiency of pulmonary nodule diagnosis are improved, and the usability and robustness are high.

The invention discloses a cross-domain target detection method based on multilayer feature alignment. The method includes: training a detector on a source domain data set with a frame label through adeep convolutional neural network; then, taking the trained detector as a pre-training model, and carrying out feature extraction on the pictures of the source domain and the target domain without theframe label through a deep convolutional neural network VGG-16 to enable the source domain and the target domain to share feature parameters; secondly, designing a domain classifier, taking the extracted feature layers of the multi-layer source domain and the target domain as the input of the domain classifier, and judging whether the feature layers are from the source domain or the target domain; and aligning the feature distribution of the source domain and the feature distribution of the target domain through the training mode of the generative adversarial network, so as to reduce the datadeviation between the two domains; and finally, carrying out joint training on the detector and the discriminator to obtain a final model. According to the invention, the knowledge of the source domain is migrated to the target domain, and the detection precision of the target domain data without frame annotation is improved.

The embodiment of the invention provides a named entity recognition model training method and a named entity recognition method. According to the invention, the method comprises the steps: employing atraining set for training a BERT-CRF model to obtain a named entity recognition model trained in this round; marking the to-be-identified data set by using the named entity identification model trained in this round to obtain a weakly marked to-be-identified data set; selecting one part from the weakly marked to-be-identified data set and taking the part and the initial training set as a new training data set to continuously carry out the next round of training on the named entity identification model; therefore, the named entity recognition model uses the to-be-recognized data set to adjustthe model before recognizing the to-be-recognized data set, so the named entity recognition model has better generalization ability, and finally the recognition effect of the model on the to-be-recognized data set is improved.

The general demodulation (DEMOD) method includes steps: under precondition of accorded with sample theorem, full phases FFT, and required corrected sampling points to select A/D conversion for the sampling frequency (FR); selecting sequence needed by full phases from the sampling sequences, and then based on FR spectrum correction (FSC) technique to select the correction sequence; carrying out FFT in full phases for the selected sequence, and meanwhile based on the selected FSC technique to carry out FSC so as to accomplish demodulation (MOD); using planisphere-mapping to accomplish decoding. Since DEMOD is based on full phases FFT, the algorithm for correcting and accurate estimating FR, phase, and amplitude is simple so as to simplify DEMOD circuit. Features are high resolution for estimating FR and phase, and low error rate. The invention is suitable to multiple MOD modes including different distributions of planisphere so as to be in favor of generalization of demodulator.

The invention discloses a bearing-free asynchronous motor RBF neural network self-adaptive inverse decoupling control and parameter identification method. An SVPWM module, a voltage inverter, a bearing-free asynchronous motor and a load of the bearing-free asynchronous motor form a whole serving as a composite controlled object. Two radial basis function neural networks are adopted to achieve inverse control and parameter identification conducted on the composite controlled object. A self-adaptive inverse controller is formed by using an RBF neural network through learning, and is serially connected in front of the composite controlled object, errors of a feedback signal and a given signal are input into an inverse controller, and accordingly closed-loop control is formed, then a self-adaptive parameter identifier is formed by using one RBF neural network through learning and identifies output quantity speed and displacement of the composite controlled object, speed-less and displacement-free sensor control is achieved, online learning of an estimation signal is aided by means of a learning algorithm, and non-linear dynamic decoupling control of the bearing-free asynchronous motor is achieved. The bearing-free asynchronous motor RBF neural network self-adaptive inverse decoupling control and parameter identification method is high in control speed and higher in identification accuracy, and a control system is excellent.

The invention belongs to the technical field of computer image recognition, and specifically relates to a pedestrian re-identification method based on attitude normalized image generation. The methodcomprises steps of: predicting a pedestrian average attitude and attribute features; constructing, training and testing an attitude normalized image generation model and generating eight pedestrian images having different attitudes; constructing, training and testing pedestrian re-identification feature extraction network to obtain a pedestrian re-identification feature; and finally, performing the pedestrian re-identification feature fusion, and obtaining the features of a pedestrian target to be detected and all candidate pedestrian targets. The method of the invention has the advantages ofhigh speed, high precision, good robustness, good generalization ability and good expandability, and is very suitable for practical applications such as video pedestrian monitoring and video pedestrian information retrieval.

The invention discloses a magnetic resonance imaging (MRI) based brain disease individual prediction method and a magnetic resonance imaging (MRI) based brain disease individual prediction system. The method comprises the following steps: 1: obtaining the MRI of the brain of a patient with mental diseases; 2: carrying out denoising and dimension reduction treatment on the MRI of the brain of the patient; 3: carrying out feature selection by utilizing a ReliefF algorithm; 4: adaptively obtaining a spatial brain area by using a spatial cluster analysis method; 5: removing redundant features by utilizing a correlation-based feature selection algorithm, thus obtaining an optimal feature subset; 6: carrying out multiple linear regression analysis based on the optimal feature subset to recognize potential biomarkers. The method has the beneficial effects that the embodiment of the invention integrates various machine learning methods and can rapidly and conveniently achieve quantitative and individual accurate prediction of the interest features of mental diseases, such as clinical indexes, based on various image data in different mode types, thus being beneficial to understanding the brain structures, function abnormity and potential pathogenesis of the diseases.

The present invention provides a recommendation model training method and training apparatus, which can take the behavior of the user after the recommendation result is displayed as recommendation feedback, and can effectively alleviate the sparseness of the acceptation amount in the training data compared with the recommended amount and the imbalance of the proportion of the positive and negative samples. The recommendation model training method comprises: obtaining training data of a recommendation model, wherein the training data is the data generated in the latest time period, and the latest time period has a predefined time length; marking the training data according to the predetermined training data marking rule to obtain the latest marked data, wherein the latest marked data comprises the user feedback after the recommendation result of the latest time period is displayed, and the user feedback is determined according to the user behavior after the recommendation result is displayed; and training the latest marked data to obtain the recommendation model.

The invention discloses a single classifier anomaly detection method based on a multilayer random neural network. The the method comprises: only inputting a training data set of a normal class; through multilayer ELM-AE autoencoder and decoding processing, input sample data obtaining a reconstructed characteristic value; inputting the reconstructed characteristic value into the last layer of ELM to obtain actual output; sorting the obtained distance error vectors of the actual output and output tags from large to small, and determining a threshold for separating a normal class from an abnormalclass according to a set threshold parameter; and finally, inputting the test data into the multi-layer random neural network single classification abnormity detection model, and testing the recognition effect of the model. According to the method, main information is extracted more quickly and efficiently, dimensionality reduction is carried out, and then recognition and classification are carried out. And the speed is higher, the accuracy is higher, and the generalization performance is better. The method is not only suitable for small data sets, but also suitable for high-dimensional largedata sets, and has universality. And the method has important significance for practical application in future.

The invention discloses an ancient font classification method based on a convolutional neural network. According to the method, firstly, an ancient font category image data set is crawled by using a crawler technology; through data expansion, training set samples tend to be balanced; graying processing is carried out on the balanced training set sample and setting an image size to a target image size; histogram equalization processing is performed on the sample set, isolated noise points are removed in the image through an N8 connected noise reduction algorithm, and finally binarization processing is performed on the image based on a fuzzy set theory and by using a Shannon entropy function, so that detail features of the image are well reserved; based on the objective function of the classification task. The center loss function and the traditional cross entropy loss function are matched for use. The inter-class distance is increased. The intra-class distance is reduced. The distinguishing capability of features is improved to a certain extent, preprocessed images are trained through a pre-defined network model, and the accuracy of a classification result is evaluated through a confusion matrix. According to the method. The preprocessing effect on the degraded ancient font image is remarkable, and a more accurate ancient font classification effect is achieved by optimizing parameter setting and utilizing appropriate training skills to train the convolutional neural network model.