90 results about "Network model" patented technology

A computerized neural-network based mechanism for providing an intermediary configured for intervening in searches is described. Corresponding methods, computer-readable media, systems, devices, and apparatuses are also contemplated. The neural network can include a multi-headed attention layer. The intermediary may be, in some embodiments, a human “man in the middle” mechanism invoked where there is low confidence that pre-existing categories map to a user's search string. The mechanism provides a specially configured interface adapted to enable a search specialist to quickly select one or more categories that match or are otherwise associated with the search query from a set of acceptable categories. Received outputs and detected user behaviors are utilized to update a neural network model.

The invention relates to an improved particle swarm-optimized neural network-based transformer fault diagnosis method. The method includes the following steps that: related data of dissolved gases in transformer oil and transformer fault information are obtained so as to be adopted as sample data, and a drop half-normal distribution scoring model is adopted to pre-estimate the data of the dissolved gases in the transformer oil; the network structure of a neural network is determined; and the parameters of the neural network are optimized by using an improved particle swarm algorithm; pre-estimated sample data are adopted to train the parameter-optimized neural network, so that a final neural network model can be obtained; and the neural network model is adopted to process transformer data to be evaluated, so that the fault type of a transformer can be obtained through diagnosis. With the method of the invention adopted, the interference of original data redundancy information can be reduced, and the validity of data evaluation can be improved; and convergence speed in the training of the neural network can be increased, and the search ability of parameter optimization can be improved, and the accuracy and reliability of transformer fault diagnosis can be improved finally.

The embodiment of the invention discloses an eye fundus image recognition method and device, electronic equipment and a storage medium. The method comprises the steps of obtaining a to-be-identified fundus image, and performing grid partitioning processing on the to-be-identified fundus image to form a plurality of to-be-identified partitioned grids; Inputting each to-be-identified block grid of the to-be-identified fundus image into a pre-trained neural network model to determine a focus state of the to-be-identified block grid; And determining the focus state of the to-be-identified fundus image according to the focus state and the position of each to-be-identified block grid in the to-be-identified fundus image. According to the technical scheme provided by the embodiment of the invention, the number of required fundus image samples and the data calculation amount in the model training process are reduced while the focus recognition accuracy and multi-type focus recognition are ensured.

The invention provides a behavior recognition method and system based on an attention mechanism double-flow network, and belongs to the technical field of behavior recognition, and the method comprises the steps: dividing an obtained whole video segment into a plurality of video segments with the same length, extracting an RGB image and an optical flow gray-scale image of each frame of each videosegment, and carrying out the preprocessing of the RGB images and the optical flow gray-scale images; carrying out random sampling on the preprocessed image to obtain an RGB image and an optical flowgrayscale image of each video clip; extracting appearance features and time dynamic features of the sampled images by using a double-flow network model introducing an attention mechanism, fusing the appearance features and the time dynamic features according to the types of a time domain network and a space domain network respectively, and performing weighted fusion on a fusion result of the timedomain network and a fusion result of the space domain network to obtain an identification result of the whole video. According to the invention, the video data can be fully utilized, the local key features of the video frame image can be better extracted, the foreground area where the action occurs is highlighted, the influence of irrelevant information in the background environment is inhibited,and the behavior recognition accuracy is improved.

The invention discloses an emotion recognition method and system based on a deep learning model and a long-short memory network, The method comprises the following steps: data preprocessing and data set partitioning of EEG signals are performed to construct a network model, wherein the network model comprises a picture reconstruction model composed of a variational encoder and an emotion recognition model composed of a long-short memory network; the network model comprises an image reconstruction model composed of a variational encoder and a short-long memory network; The objective function isconstructed according to the network model. The network model is trained by training set, and the objective function is optimized by Adam optimizer in neural network, and the trained network model isobtained. Using the cross-test set to cross-test the trained network model, determining the super-parameters of the network model, and obtaining the final network model; and using the final network model to visualize the seed data and perform emotion recognition. The invention relies on the data artificial intelligence method to learn the collected EEG signal space and time complex structure, reduce the subjective factors in the prediction, and improve the prediction accuracy.

The invention provides a malicious software API (Application Program Interface) call sequence detection method based on graph convolution. The method comprises the following steps: acquiring and recording API call sequence information of processes and sub-processes when a large number of software samples run; performing vectorization processing on the API calling sequence information; extracting aparameter relationship, a dependency relationship and a sequence relationship of the API function; establishing an API call graph; inputting the API call graph into a graph convolutional neural network for training to obtain a malicious software detection network model; collecting API calling sequence information of processes and sub-processes when the executable file to be detected runs; constructing an API call graph of the executable file to be detected, then inputting the API call graph of the executable file to be detected into the malicious software detection network model, If the output result of the malicious software detection network model is 1, indicating that the judgment result is malicious software; If the output result of the malicious software detection network model is 0,indicating that the judgment result is normal software.

The invention is suitable for the technical field of ice shape prediction, and provides a deep belief network-based airfoil profile icing ice shape prediction method and device. The method comprises the following steps: constructing and training a Fourier coefficient deep belief network model and an upper and lower limit deep belief network model in advance; performing data normalization on the icing condition to be predicted to obtain a normalized icing condition; inputting the normalized icing condition into the Fourier coefficient depth confidence network model and the upper and lower limitdepth confidence network model; substituting ai, bi, xiu and xil into an ice-shaped curve Fourier series expansion formula to obtain a wing-shaped icing ice-shaped curve; the Fourier coefficient deepbelief network model and the upper and lower limit deep belief network model are composed of a plurality of restricted Boltzmann machines and a BP neural network layer. According to the method, the network training time is greatly reduced, the network prediction precision is improved, and the technical problems that gradient disappearance and local minimization are likely to happen to a pure BP neural network are solved.

The invention discloses a method, system and device for compiling an AI chip and a storage medium, and the method comprises the steps of carrying out the quantification of a network model of the AI chip, and carrying out the precision adjustment of the quantified network model based on the weight of the network model; allocating calculation of the convolutional neural network to different MAC calculation units of the AI chip in the network model after precision adjustment so as to optimize a calculation process of the network model; and acquiring hardware architecture parameters, and generating a continuous calculation flow according to the hardware architecture parameters and the network model. According to the invention, the AI chip is quantified and the precision is adjusted, so that the performance is more perfect; the calculation of the network is optimized to the maximum extent, the calculation steps are reduced, and the utilization rate of the MAC calculation unit is maximized;the utilization rate of the AI equipment is improved by generating the calculation flow capable of enabling the calculation equipment to be always in the busy state in the network reasoning process.

The invention discloses a track circuit fault automatic diagnosis method based on a deep convolutional neural network, and the method comprises the steps: firstly obtaining the real-time change data of a track circuit, building a track circuit fault type data set, and dividing the data set into a training sample and a test sample; secondly, establishing a one-dimensional deep convolutional neuralnetwork, inputting the training set sample to train a network model, and evaluating the accuracy of the network model through a test sample to obtain a trained one-dimensional deep convolutional neural network model; and finally, inputting the real-time change data of the track circuit into the trained deep convolutional neural network model, establishing a mapping relationship between input and output through a neural network, outputting a fault type, and positioning a fault position. According to the method, the fault diagnosis efficiency and accuracy of the track circuit can be improved, field maintenance personnel can quickly process faults, and the maintenance and repair efficiency and the intelligent level of the track circuit are improved.

The invention discloses a product quality control method based on a process network model and a machine learning algorithm. According to the method, product quality control is realized in a mode of fusing a product process complex network and a machine learning algorithm, namely, a quality transmission complex network is established based on a product process mechanism, a machine learning XGBoost model is established based on the quality transmission network and sample data and trained, and an SHAP algorithm model is established to analyze an XGBoost training result. Key process parameters influencing quality are quantitatively mined, and the linkage effect among the process parameters is accurately calculated. The method has the advantages that a product process mechanism and a big data analysis method are effectively combined, so that the problems of high dimension, strong nonlinearity and non-uniform sample distribution of industrial actual production data, which cannot be solved by a common statistical analysis method, are effectively solved, and the defect that a traditional product quality control method carries out qualitative description or only carries out product quality control for a single independent factor is overcome; and the method can comprehensively and quantitatively analyze the influence of complex process factors in the whole process of product processing on the final product quality and accurately calculate the linkage influence among the process factors to form a quality identification model based on process mechanism and data dual drive, and is large in processing data volume, high in speed and accurate in evaluation result.

The invention discloses a structure optimization design method for accelerating by using a graph convolutional neural network structure. The method can solve the problems that a traditional Ground Stream algorithm is high in calculation complexity and large in time expenditure in structure optimization, the method comprises a model obtaining part and a model using part, and the model obtaining main process comprises the steps that 1, input sample data set standardization processing is carried out; 2, generating (10000 groups) optimized topological structure diagrams and standardized input diagram random distribution as a training set, a verification set and a test set by using a Ground Stream algorithm; 3, constructing a graph neural network model; and 4, training a graph neural network model, and storing the model and parameters. And finally, the network model is used for calculation, testing and evaluation, so that quick calculation of a final optimized structure is realized, the calculation cost is reduced, and the time overhead is improved.

The invention provides a neural network quantification method based on parameter norms. The method comprises the steps that for a given pre-trained neural network parameter model, a quantification center is divided by performing statistical analysis on values of parameters of needed quantification layers; according to the selected quantification center, quantification loss of the parameters of each corresponding quantification layer is calculated; the quantification loss and classification error loss of the neural network training parameter model are added to serve as total loss, back propagation optimization is performed, and meanwhile the quantification center is updated during optimization; and after training is over, quantification operation is performed on the corresponding layers according to the quantification center, and a compression model after quantification is obtained. Through the method, a weight center can be divided, and the neural network model is quantified by exerting simple quantification loss and using an optimizer the same as that in a traditional method, so that the compression model of the original model is obtained, and the network storage volume and operation complexity are reduced.

The invention belongs to the technical field of digital forensics, and particularly relates to a malicious code fragment intelligent forensics method and system, and the method comprises the steps: constructing a code fragment training set and a code fragment test set for training and testing through extracting the underlying data features of a storage medium; training the set full-connection neural network model by using the data in the code fragment training set, the input being a feature vector, and the output being a normal or malicious prediction result; for the code fragment test set, performing test output by utilizing the trained full-connection neural network model to judge whether model input is a malicious code fragment; and performing feature extraction on the target code snippets, and inputting the target code snippets into a fully-connected neural network model generated through training and testing to obtain an intelligent malicious code recognition result of the targetcode snippets. According to the method, malicious code fragments in storage media such as computers, mobile phones and tablets and evidence containers such as RAW, E01 and AFF can be recognized, and the method has a good application prospect in the field of digital evidence collection such as crime event evidence underlying data automatic analysis.

The invention discloses an accurate prediction method and system for horizontal displacement of a concrete faced rockfill dam. The accurate prediction method comprises: obtaining and preprocessing historical measured data of the concrete faced rockfill dam to be measured; classifying environmental factors influencing the horizontal displacement of the concrete faced rockfill dam, and selecting aninput variable of a multiple linear regression model according to the correlation between each influence factor and the horizontal displacement; establishing a concrete faced rockfill dam horizontal displacement prediction multiple linear regression model considering rockfill material delay response and cyclic loading and unloading deformation characteristics, and obtaining a preliminary prediction value of horizontal displacement; and establishing a statistical optimization neural network model and carrying out optimization training, and taking a dependent variable of the multiple linear regression model and the preliminary prediction value as inputs of the statistical optimization neural network model to obtain a horizontal displacement prediction value of the concrete faced rockfill dam. The horizontal displacement of the concrete faced rockfill dam can be accurately predicted, and safe operation of the dam and accessory structures of the dam is guaranteed.

The embodiment of the invention discloses a base station planning method and a base station planning device, relates to the technical field of communication, and can predict the base station planningprobability by adopting a BP neural network algorithm for performance data of grids so as to timely realize base station planning. The method comprises the following steps: carrying out rasterizationprocessing on an area to be planned to generate at least one grid; Inputting the obtained performance data of the predetermined time into a preset model to generate a first probability of the grid planning base station after the predetermined time; Wherein the preset model is generated by training the training sample set by the BP neural network model; Wherein the training sample set comprises training samples, the training samples comprise historical performance data of grids in a planning time period and a second probability of planning base stations in the grids after the planning time period, and the second probability represents that the grids need or do not need to plan the base stations after the planning time period; And if the first probability is greater than a preset probabilitythreshold, determining to plan the base station in the grid. The embodiment of the invention is applied to a communication system.

The embodiment of the invention provides a surgery robot system. The system comprises an image collecting device for collecting surgery environment information to obtain color image information, a positioning device for obtaining pose information of the image collecting device, and a recognizing device which is in communication connection with the image collecting device and the positioning deviceand is used for obtaining feature information according to the color image information, wherein the pose information is used for representing the position and pose of the image collecting device inside a patient body, the recognizing device is further used for adopting the feature information and/or the pose information as input on the basis of a neural network model and outputting labelled information. According to the surgery robot system, on the basis of the neural network model, the content in the view of the image collecting device and the relative position where the current image collecting device is located can be rapidly output in real time, and then a medical worker can be helped to accurately perform operation and observation when surgery is performed on a patient.

The invention discloses a false quantity recognition model generation method, a false quantity recognition method and a computing device. The false quantity recognition model generation method comprises the steps that role data corresponding to all role identifiers is acquired; data statistical analysis is performed on all the role data to generate first role statistical data corresponding to allIP addresses which are logged in and second role statistical data corresponding to all devices which are logged in; a false quantity IP address ratio and a false quantity device ratio corresponding toall the role data are calculated; all the role data is associated with the corresponding false quantity IP address ratio and false quantity device ratio to generate corresponding role extension data,clustering processing is performed on all the role extension data to acquire an optimal quantity of role clusters, and role cluster tags of all the role clusters are generated; role tags of all the role extension data are generated; and according to all the role data and the role tags associated with the role extension data corresponding to the role data, a neural network model constructed in advance is trained to generate a false quantity recognition model.

The invention discloses a nuclear magnetic resonance image automatic classification method and device based on an MDCLSTM-LdenseNet network, and belongs to the field of medical image processing. The method comprises the following steps: obtaining nuclear magnetic resonance images of different types of subjects; preprocessing the nuclear magnetic resonance image to obtain a training data set, a verification data set and a test data set; training is carried out on the constructed MDCLSTM-LDenseNet network model, so that the MDCLSTM-LDenseNet network model can be obtained; inputting the test dataset into the trained MDCLSTM-LDenseNet network model to carry out a test; obtaining classification results of the nuclear magnetic resonance images of the three types of subjects with normal cognition, mild cognitive impairment and Alzheimer's disease and classification accuracy of the MDCLSTM-LdenseNet network model; according to the method and the device, used data sets are strictly divided; data leakage is avoided; the method is advantaged in that the training data set, the verification data set and the test data set are not intersected, reliability of classification results of normal cognition, mild cognitive impairment and Alzheimer's disease is guaranteed, classification accuracy is higher, the missed diagnosis rate and the misdiagnosis rate are lower, and doctors can be more effectively and reliably assisted to diagnose Alzheimer's disease.

The invention relates to an internal threat detection method and system based on a twin network, and the method comprises the steps: S1, obtaining an internal threat test data set, carrying out the preprocessing, constructing a sample set, and dividing the sample set into a training set, a to-be-detected sample set, and a standard sample set; s2, randomly selecting a sample pair and a category label thereof from the training set to form an input triple, inputting the triple into a twin network model, calculating an Euclidean distance of two obtained two-dimensional sample vectors, and judging whether the two two-dimensional sample vectors belong to the same category or not according to a preset threshold value; and S3, respectively taking one sample from the to-be-detected sample set and the standard sample set, inputting the samples into the trained twin network model, and judging whether the two samples belong to the same category so as to determine the category of the to-be-detected sample. The detection method provided by the invention comprehensively analyzes the user behavior, realizes more accurate internal threat detection, is insensitive to whether the data set is balanced or not, and improves the intelligence of the system.

The invention discloses a sample generation method and device. The method comprises the steps: inputting data in a pre-collected training set into a neural network model, obtaining a sample generationmodel, and enabling the neural network model to be a neural network model employing an attention mechanism; Inputting pre-collected data in the test set into the sample generation model to generate asample. According to the invention, a neural network model of an attention mechanism is adopted; the neural network model can endow characters in the input data with different weights; Therefore, data information in the input data can be learned more selectively, data with high correlation with the current output data in the input data can be found, a high-precision deep learning model can be finally obtained, training loss is low, and samples generated by the high-precision deep learning model are inevitably higher in legality and diversity.