913 results about "Prediction probability" patented technology

The invention relates to an automatic detection system for a pulmonary nodule in a chest CT (Computed Tomography) image. The system provides improvements for the problems of large calculated amount of computer aided software, inaccurate prediction and few prediction varieties. The improvement of the invention comprises the steps of acquiring a CT image; segmenting a pulmonary tissue; detecting a suspected nodular lesion area in the pulmonary tissue; classifying nidi based on a nidus classification model of deep learning; and outputting an image mark and a diagnosis report. The system has high nodule detection rate and relatively low false positive rate, acquires an accurate locating, quantitative and qualitative result of a nodular lesion and a prediction probability thereof. The end-to-end (from a CT machine end to a doctor end) nodular lesion screening is truly realized, the accuracy and operability demands of doctors are satisfied and the system has wide mar5ket application prospects.

The invention provides an artificial intelligence-based object pushing method and apparatus. The method comprises the steps of inputting historical click behaviors of a target user to a built user model for performing learning to obtain multi-dimensional preference eigenvectors of a target user; obtaining eigenvectors of all to-be-pushed objects; inputting the eigenvectors of the to-be-pushed objects and the multi-dimensional preference eigenvectors to a trained deep neural network model for performing prediction to obtain prediction probabilities of the to-be-pushed objects; and pushing the to-be-pushed objects to the target user according to the obtained prediction probabilities. According to the method and the apparatus, the preferences of the target user are obtained through the user model built by a neural network, and the probability of possibly purchasing to-be-pushed group orders by the target user is obtained based on the deep neural network and the preferences of the target user, so that the pushing is more effective; and preference features are selected through training by the user model, so that a large amount of manpower does not need to be consumed for selection and the pushing efficiency is improved.

The invention relates to a tag noise correction based crowd-sourced tagging data quality improvement method. The method comprises the following steps: running a tag integration algorithm in an initial crowd-sourced tagging data set to form a data set after tag integration, and estimating tagger quality and integrated tag quality information of samples in the process; performing multi-round K-fold cross validation by utilizing the data set after tag integration, and constructing a high-quality data set; determining a tag noise set in combination with the tagger quality and the tag quality of the samples by utilizing a prediction probability of a class tag of each sample in the multi-round K-fold cross validation process; and training a classification model by utilizing the high-quality data set generated in the multi-round K-fold cross validation process, and performing prediction and replacement on the class tag of each sample in the tag noise data set by using the model. With the tag noise correction method, the quantity of potential noise tag samples in the data set after original tag integration is reduced, so that the data quality is improved.

The invention relates to a reservoir prediction method based on a bidirectional recurrent neural network. The reservoir prediction method includes: generating well seismic data, generating well markeddata, generating a marked dataset, building and training the bidirectional recurrent neural network, storing a trained model, extracting seismic data corresponding to all geographic positions in a region, and performing prediction on the seismic data to acquire probability distribution of reservoirs below the whole region. A one-to-one corresponding relation between each sampling point and reservoirs and non-reservoirs is established, wherein input of each time step is n-dimension seismic data of each sampling point while output of the same is corresponding reservoir or non-reservoir marks. An optimal model is acquired by training marked samples and adjusting hyper-parameters, in this way, unmarked data can be predicted to finally generate a predicted seismic body of same dimension, and value of each sampling point is prediction probability between 0 and 1, so that the method has good effect on reservoir prediction.

The invention discloses an identification method of a harassment number. The identification method of the harassment number comprises the steps of selecting a plurality of harassment numbers and non-harassment numbers which are confirmed; calculating communication behavior indexes of the harassment numbers and non-harassment numbers within a period of time; then forming a training sample set by using the harassment numbers and non-harassment numbers as well as the communication behavior indexes thereof so as to build a random forest classification model, wherein the input of the random forest classification model is the communication behavior index of each user number and the output thereof is a prediction probability of judging each user number as a harassment number or non-harassment number by all the decision-making trees; and inputting the communication behavior index of a to-be-identified number within a period of time into the random forest classification model, calculating the prediction probability of judging the to-be-identified number as a harassment number or non-harassment number by all the decision-making trees, and accordingly determining whether the to-be-identified number is a harassment number. The identification method of the harassment number is belongs to the technical field of network communication, can efficiently identify harass numbers from massive traffic data of the present network by making full use of calling features of calling and called numbers.

The invention discloses a whole-process optimizing method of a multi-stage batch production process. The whole-process optimizing method of the multi-stage batch production process comprises the steps of collecting and preprocessing a plurality of batches of production data of a multi-stage batch production process, wherein the production data comprises procedure parameters and product quality attribute data; establishing a partial least square (PLS) model between the procedure parameters and the product quality attribute data by means of the PLS method; and determining an optimizing objective, combining with the PLS model and calculating Bayesian posterior prediction probability, coinciding with the optimizing objective, of product quality attribute, and selecting largest procedure parameter combination of the Bayesian posterior prediction probability as the optimizing result. According to the whole-process optimizing method of the multi-stage batch production process, optimizing technological operation coinciding with established quality objectives can be provided for every stage of the multi-stage batch production process, and finally stability and improvement of product quality are achieved.

The invention discloses a text recognition method under a natural scene on the basis of spatial transformation. The method comprises the following steps: firstly, obtaining the text contents of a text image in a training image set, and training network models including a datum point positioning network, an image preprocessing network, an image encoder network, a feature encoder network and the like; then, utilizing the network models obtained by training to carry out spatial transformation on an image in a set of images to be identified to obtain the transformed image to be identified; and then, calculating the feature vector and a prediction probability sequence of the transformed image to be identified, and finally obtaining an image recognition result. The method is high in text recognition accuracy and can overcome the influence of adverse factors including irregular text arrangement and the like.

The invention discloses a credit card anti-fraud prediction method based on a dual-mode network diagram mining algorithm. The method concretely comprises the following steps that the original data ofthe credit card applicant is acquired and the original data are converted into the diagram data; the nodes, the edges, the attributes of the nodes and the attributes of the edges required for constructing a dual-mode network model are selected out of the diagram data through screening; the dual-mode network model is constructed; a network risk characteristic model is constructed and the probability of network fraud is acquired; the probability of personal fraud is acquired; and the probability of network fraud and the probability of personal fraud are integrated so as to obtain the fraud prediction probability of the credit card applicant. The multidimensional data information related to the applicant is collected, the credit card application field data knowledge map is constructed, the dual-mode network model capable of reflecting the correlation between the clients is acquired and the influence of the individual and group risk on the applicant fraud probability can be accurately integrated so that the risk of identity forgery, group fraud and group attack can be effectively reduced, and the financial anti-fraud risk control capacity can be enhanced.

The invention provides a face attribute recognition method based on a combined depth neural network. The method comprises the following steps: S1, performing area extraction on an inputted face image;S2, carrying out positioning on a facial important part and extracting a related area based on the area extraction result to obtain facial area features and facial key part features; S3, marking a real age and a gender of each face by using the area extraction result as a training sample, carrying out training based on a deep neural network, and outputting an age estimation value and a gender recognition result; and S4, on the basis of the facial area features outputted by the S2 and local area features of eyes and mouths, outputting expression and wearing prediction probabilities respectively according to a random forest algorithm, adding probabilities of corresponding attributes for different attributes, and outputting facial expression and wearing attribute results. In addition, the invention also provides a face attribute recognition apparatus based on a combined depth neural network.

The invention discloses an automatic text classification method based on BERT and feature fusion. The method comprises the following steps: firstly, cleaning text data, realizing conversion from a text to a dynamic word vector through BERT, extracting features of the text by utilizing CNN and BiLSTM, and respectively transmitting a word vector sequence output by the BERT to a CNN network and a BiLSTM network; then, splicing the output of a CNN network and the output of a BiLSTM network together, carrying out feature fusion, and finally, outputting a final prediction probability vector througha full connection layer and a softmax layer. The method is suitable for the general supervised text label prediction problem, and can effectively improve the prediction accuracy of the text data labels with prominent sequence information and local features.

A system and method for actively selecting and labeling images for semantic segmentation are disclosed. A particular embodiment includes: receiving image data from an image generating device; performing semantic segmentation or other object detection on the received image data to identify and label objects in the image data and produce semantic label image data; determining the quality of the semantic label image data based on prediction probabilities associated with regions or portions of the image; and identifying a region or portion of the image for manual labeling if an associated prediction probability is below a pre-determined threshold.

The invention provides an unsupervised field adaptation method and system based on adversarial learning and a medium, and the method comprises the steps: a feature extraction step: employing a featureextraction network to extract the features of images in a source field and a target field, and obtaining the image features of the source field and the image features of the target field; a categoryprediction step: predicting the probability that the image belongs to each category according to the obtained image feature of the source field and the image feature of the target field, and obtaininga category prediction probability; and a field discrimination step: predicting the probability that the image feature comes from the source field and the target field through a field discrimination network according to the obtained image feature of the source field and the image feature of the target field, and obtaining a field prediction probability. According to the method, the characteristicsthat the image extraction fields in the source field and the target field are invariable and have relatively high discrimination capability can be realized, so that the unsupervised field adaptationis realized.

The invention relates to a medical image imaging method and device, computer equipment and a storage medium. The method comprises the steps of obtaining an original medical image and segmentation information; performing gray correction and image enhancement on the original medical image to obtain a to-be-segmented image; segmenting a plurality of regions of interest from the to-be-segmented imagebased on the segmentation information; inputting each region of interest into the combined prediction model to obtain a prediction probability corresponding to each region of interest; and generatinga target medical image according to each prediction probability. By adopting the method, the time can be shortened, and no side effect is generated to patients.

The invention provides a relevance vector machine-based multi-class data classifying method, which mainly solves the problem that the traditional multi-class data classifying method cannot integrally solve classifying face parameters and needs proximate calculation. The relevance vector machine-based multi-class data classifying method comprises a realizing process comprising the following steps of: partitioning a plurality of multi-class data sets and carrying out a normalizing pretreatment; determining a kernel function type and kernel parameters; setting basic parameters; calculating the classifying face parameters; calculating lower bounds of logarithms and solving variant values of the lower bounds of the logarithms and adding 1 to an iterative number; if the variant values of the lower bounds of the logarithms are converged or the iterative number reaches iterating times, finishing updating the classifying face parameters, and otherwise, continuing to updating; and obtaining a prediction probability matrix according to the updated classifying face parameters, wherein column numbers corresponding to a maximum value of each row of the matrix compose classifying classes for testing the data sets, and samples which have the prediction probability less than a false-alarm probability and the detection probability corresponding to a false-alarm probability value set in a curve are rejected. The relevance vector machine-based multi-class data classifying method has the advantages of obtaining classification which is comparable to that of an SVM (Support Vector Machine) by using less relevant vectors and rejecting performance and can be used for target recognition.

The invention provides a recurrent neural network attention model-based pedestrian attribute recognition network and technology. The pedestrian attribute recognition network comprises a first convolutional neural network, a recurrent neural network and a second convolutional neural network, wherein the first convolutional neural network is used for extracting a whole body image feature of a pedestrian by taking an original body image of the pedestrian as an input; the recurrent neural network is used for outputting an attention heat map of an attribute group concerned at the current moment andlocally highlighted pedestrian features by taking the whole body image feature of the pedestrian as a first input and taking an attention heat map of an attribute group concerned at the last moment as a second input; and the second convolutional neural network is used for outputting an attribute prediction probability of the currently concerned group by taking the locally highlighted pedestrian feature as an input. According to the network and technology, a recurrent neural network attention model is utilized to mine an association relationship of pedestrian attribute area space positions soas to highlight positions of areas corresponding to attributes in images, so that higher pedestrian attribute recognition precision is realized.

The invention aims to solve problems of poor effect, small quantity of training samples, large difference among samples of a traditional extracted feature resorting method due to large difficulty in melanoma skin lesion segmentation and provides a multi-scale deep learning recognition system based on expansion and convolution. The system includes performing data enhancement and normalization processing on training samples and training an extracted expansion and convolution based multi-scale feature learning neural network, performing multi-threshold segmentation based on an obtained predictionprobability map, and thus implementing segmentation of a melanoma skin lesion image. Finally, the segmentation accuracy is improved.

The invention relates to a target detection model training method and device, a computer readable storage medium and computer equipment, and the method comprises the steps: obtaining a feature map ofa sample image during training, and determining an initial detection box in the feature map according to a preset rotation angle, a preset scale and a preset target aspect ratio; adjusting the position of each initial detection frame to obtain the position information of the prediction detection frame, and adjusting the network parameters of the regression network according to the position information and the real position information in the annotation information of the sample image; predicting the prediction probability of the target corresponding to each preset category according to the target detection area determined by the position information of the prediction detection frame; and adjusting the network parameters of the classification network according to the real category information in the annotation information of the sample image and the prediction probability, and then obtaining a target detection model used for performing target detection on the image. According to the scheme provided by the invention, the target detection model can identify the rotation angle of the target in the image, and the positioned target detection frame is more accurate.

The invention discloses a medical image interpretation method and device, computer equipment and a storage medium, and the method comprises the steps: obtaining an image analysis request which comprises a target medical image; adopting a pre-trained image recognition model to recognize the target medical image, and obtaining a feature map output by the last convolutional layer in the image recognition model; based on the feature map, obtaining a prediction probability value corresponding to each original focus category output by the image recognition model; determining the original lesion category with the maximum prediction probability value as a target lesion category; obtaining a map weight corresponding to a target lesion category, carrying out classified activation mapping on a characteristic map and the map weight by adopting an activation mapping formula, obtaining a thermodynamic mapping map, superposing the thermodynamic mapping map and a target medical image, generating a target thermodynamic map, and improving an image recognition rate.

The invention relates to a probability route improving method in a delay-tolerance mobile sensor network based on the node activeness. The method adds node activeness factors into prediction of the transmission probability to provide a calculating formula of the self node activeness NA, the NA and a transmission prediction probability value of an original PROPHET algorithm are weighed through a weight alpha, a new transmission prediction probability value is calculated according to node meeting history information and node activeness in the network, whether information is transmitted is decided by comparing transmission prediction probability values of nodes, and buffer memory is managed through a time to live TTL discarding strategy. Simulation results show that a probability estimating method of an NAPR algorithm is more comprehensive than a PROPHET algorithm, fewer copies are generated, the NAPR algorithm reduces the node overhead rate even though the average delay of the information is slightly increased, and the delivery rate of message is improved. In future work, the probability route improving method can be further used for solving the jam problem in the network.

The invention discloses a response method and device of question-answer interaction, a storage medium, and a terminal. The response method of the question-answer interaction comprises the following steps: acquiring a user question; obtaining multiple answers for a user problem by using at least two ways; screening multiple answers according to a source parameter and a prediction probability of each answer, wherein the source parameter and the prediction probability of the answer are determined according to an acquisition way of the answer; and outputting an optimal answer obtained by screening. Through the technical scheme disclosed by the invention, the accuracy and the continuity of the response in the question-answer interaction process can be improved.

Methods and systems consistent with the principles of some embodiments of the present invention provide for determining a set of variables that together have strong predictive power relative to some target variable(s); identifying a subset of variables that together describe the majority of the information in the database; statistically segmenting the database; predicting unknown values in the target variables; combining prediction and VIVa; filling gaps in databases for further analysis or database completion; predicting probabilities; and outputting information based on the processed data. The outputs may relate to optimizing price to maximize return, producing a database based on postal code, identifying a trade area for sale of goods; and / or optimizing the distribution of marketing funds across various marketing channels.

The invention relates to a word forecasting method and system based on a nerve machine translation system. The word forecasting method includes the steps that parallel corpora are trained, extracting is carried out from the training result, and a phrase translation table is obtained; source language sentences in any parallel sentence pairs are subjected to matching searching, and all source language phrases contained in the source language sentences are determined; target phrase translation candidate sets corresponding to all the source language phrases respectively are found from the phrase translation table; part of obtained translations are translated according to the target phrase translation candidate sets and the nerve machine translation system, and target word sets needing to be encouraged are obtained; encouragement values of all target words in the target word sets are determined according to the attention probability and the target phrase translation candidate sets which are based on the nerve machine translation system; the prediction probability of all the target words is obtained according to the encouragement values of all the target words. The encouragement values of the target words are obtained in the mode that the phrase translation table is introduced and added into a nerve translation model, and therefore the prediction probability of the target words can be increased.

Layer processing to grow a layer structure upon a substrate surface comprises supplying a vapor mixture stream to the substrate (28) to deposit constituents, monitoring growth with an ellipsometer (12) and using its output in real-time growth control of successive pseudo-layers. A Bayesian algorithm is used to predict a probability density function for pseudo-layer growth parameters from initial surface composition, growth conditions and associated growth probabilities therewith, the function comprising discrete samples. Weights are assigned to the samples representing occurrence likelihoods based on most recent sensor output. A subset of the samples is chosen with selection likelihood weighted in favor of samples with greater weights. The subset provides a subsequent predicted probability density function and associated pseudo-layer growth parameters for growth control, and becomes a predicted probability density function for a further iteration of pseudo-layer growth.

The invention discloses an image identification model training method. The method comprises the steps of obtaining a to-be-trained image set; obtaining a first prediction probability, a second prediction probability, a third prediction probability and a fourth prediction probability through a to-be-trained image recognition model based on the to-be-trained image set; determining a target loss function according to the first prediction probability, the second prediction probability, the third prediction probability and the fourth prediction probability; and training the to-be-trained image recognition model based on the target loss function to obtain an image recognition model. The invention further discloses an image recognition method and a device. According to the method, annotation is adopted; the model is trained for the medical images of different tasks and the unlabeled medical images, the labeled images and the unlabeled images are effectively utilized, the requirement for imagelabeling is lowered, the training data volume is increased, and therefore the prediction effect of the model can be improved while labeling resources are saved.

The invention discloses a sound event detection method based on a convolutional neural network, and belongs to the technical field of audio processing. The method comprises the steps: firstly, performing primary feature extraction on an audio stream; then sending the extracted primary feature to a neural network for feature extraction and classification of sound events, and finally obtaining prediction probabilities of various sound events; and if the prediction probability of the current type of sound event exceeds a preset classification threshold, considering that a corresponding sound event exists in the current audio stream. A sound event detection model is small in parameter quantity and low in calculation complexity, so that the power consumption and the calculation complexity of the related Internet of Things equipment during sound detection processing are greatly reduced; and the detection precision equivalent to that of an existing sound event detection model is maintained. Therefore, the sound event detection method provided by the invention can be effectively applied to embedded intelligent equipment and the like.

The invention discloses a workflow unloading optimization algorithm in a mobile edge environment, comprising the following steps: step S1, constructing a workflow scheduling model under the mobile edge environment; step S2, accord to that topology structure, searching a critical path base on the local calculation amount of the workflow, respectively storing nodes and remaining node on the criticalpath into a key node queue and a remaining node queue and waiting for scheduling; step S3, after that eNB node is abstracted into a topology structure, accord to a user history moving path and combining with a prediction model, obtaining a qualified eNB set, and accord to the prediction probability, selecting an optimal unloadable eNB according to the prediction speed and the prediction direction, and providing the eNB to a user for performing task unloading; step S4, dynamically dividing that tasks accord to the scheduling strategy into local execution or unloading to the edge side for execution, and dynamically updating the critical path during the scheduling process; step S5, according to the scheduling result, assigning tasks to the devices under the mobile edge environment until thescheduling of the entire workflow is completed.

The invention discloses a diabetic retina feature grading device in an eye fundus image based on attention mechanism and feature fusion. The diabetic retina feature grading device comprises a featuredetection classification network module used for extracting first-grade diabetic retina features and second-grade diabetic retina features in the eye fundus image of an input sample, and outputting fine classification feature images extracted from the first-grade diabetic retina features and the second-grade diabetic retina features; an original image classification network module used for extracting third-grade diabetic retina features and fourth-grade diabetic retina features in the eye fundus image of the input sample, and outputting rough classification feature images extracted from the third-grade diabetic retina features and the fourth-grade diabetic retina features; an attention mechanism and feature fusion module for conducting feature fusion on the fine classification feature images output by the feature detection classification network module and the rough classification feature images output by the original image classification network module by adopting an attention mechanism, and outputting the prediction probability of the diabetic retina feature grade of the image of the input sample. The device ensures faster speed, and the classification evaluation index Kappa reaches 81.33%.

The invention discloses a text classification method based on a capsule network. The problems that in the prior art, the overall precision is not high, the applicability is not high, a large amount ofimportant information is lost in the feature extraction process, and the relation between the local part and the overall part in a text is ignored are solved. The method comprises the following steps: 1, nodes in a capsule network being capsules consisting of a group of neurons, executing complex internal calculation on input by using matrix capsules, outputting instantiation parameters from results in a matrix form, and meanwhile, outputting an activation value of each capsule is output; 2, calculating between two adjacent layers in the capsule network through an EM routing algorithm; representing a higher-dimensional concept through Gaussian cluster, and each activation capsule selecting a capsule of the next layer as a father node through an iterative routing process, so that link prediction is realized between two adjacent layers of networks; and 3, training a weight parameter of the full connection layer, and calculating the prediction probability of the real label by using a softmax activation function.

The invention relates to a video classification method, device and equipment, and a computer readable storage medium. The method comprises the steps of obtaining at least two key frame picture segments of a to-be-classified video; according to the time-space information of the at least two key frame picture segments, segment semantic vectors of the at least two key frame picture segments are acquired respectively; performing bidirectional association fusion on the at least two fragment semantic vectors to obtain a global to-be-classified semantic vector of the to-be-classified video; obtaininga prediction probability of the to-be-classified video according to the global to-be-classified semantic vector; wherein the prediction probability is used for determining a classification result ofthe to-be-classified video. By adopting the method, the video classification accuracy can be effectively improved.