296results about "Neural architectures" patented technology

Updating an artificial neural network is disclosed. A node characteristic is represented using a fixed point node characteristic parameter. A network characteristic is represented using a fixed point network characteristic parameter. The fixed point node characteristic parameter and the fixed point network characteristic parameter are processed to determine a fixed point intermediate parameter having a larger size than either the fixed point node characteristic parameter or the fixed point network characteristic parameter. A value associated with the fixed point intermediate parameter is truncated according to a system truncation schema. The artificial neural network is updated according to the truncated value.

The invention provides a relevant filtering opposite-thrust target tracking method with an adaptive scale. The method comprises: an initial position and an initial scale of a to-be-tracked target in a video frame are determined, and convolution feature graphs of different layers are extracted respectively by using the initial position as the center and using a deep convolutional neural network; for the extracted convolution feature graph of each layer, tracking is carried out by using a kernel-correlation filtering tracking method to obtain a tracking result; all tracking results are combined by using an adaptive hedging algorithm to obtain a final tracking result as a final position of the to-be-tracked target, so that the to-be-tracked target in the video frame can be localized; after obtaining of the final position of the to-be-tracked target, a final scale of the to-be-tracked target is estimated by using a scale pyramid strategy; and after obtaining of the final position and the final scale of the to-be-tracked target, a to-be-tracked target image block is extracted based on the final scale by using the final position as a center and each kernel-correlation filtering tracking method is trained again to update a coefficient and a template.

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 discloses a convolutional neural network optimization and rapid target detection method and device. The optimization method comprises the steps of obtaining a first convolutional model trained by adopting floating point type sample image data; wherein a BN layer operation in the first convolution model is located behind convolution layer operation and is used for normalizing a numerical value obtained after the convolution layer operation to a preset data range; adjusting parameters of a convolutional layer according to the parameters of the BN layer; deleting the BN layer to obtain an adjusted second convolution model; adding a quantization layer before the convolution operation of the convolution layer in the second convolution model to obtain a third convolution model; wherein the quantization layer is used for quantizing input data to integer data of a predetermined bit, and enabling a convolution layer to perform convolution operation by using the integer data of thepredetermined bit. According to the method, the problems of large model body quantity and low operation speed are solved by optimizing the convolution operation.

A method for predicting corrosion in a pipeline wherein the pipeline is subject to corrosion by the steps of obtaining data of parameters associated with the pipeline, comparing the data against previously generated data of parameters associated with the pipeline, and calculating, based on differences between the data with the previously generated data the probability of corrosion in the pipeline. The calculation can be performed using a variety of algorithms and modeling methods.

The invention discloses a remote sensing image scene classification method based on an improved residual network, which comprises the following steps of (a) collecting and/or downloading scene imagesto obtain a remote sensing scene classification data set; (b) performing enhancement processing on the images in the remote sensing scene classification data set, and randomly dividing the images in the remote sensing scene classification data set into a training set and a test set; (c) improving the residual network structure, training the improved residual network by using the images in the training set, continuously optimizing the residual network structure by training, and testing the accuracy of the residual network by using the images in the test set after the training is completed; and(d) carrying out scene classification on the remote sensing images by using the trained residual network structure. According to the method, the scene classification is carried out on the remote sensing image by utilizing the improved residual network, the network is more suitable for obtaining the remote sensing image data set with large difficulty and complex background textures by adjusting thestructure of the residual network, and the classification accuracy is improved.

The invention discloses a face recognition method and system. According to the method, the method comprises the following steps of carrying out face detection on a video stream image; after alignment,extracting a face feature vector; calculating the similarity between each frame of image and the face feature vector of the registered face feature library; establishing a tracker for each face; determining a tracker to which the human face belongs through the cross-parallel ratio of the human face position coordinate area and the weighted value of the human face similarity; and when the trackercollects face images with a preset frame number, counting a recognition result of the face tracked by the tracker, and judging a final result of face recognition according to the recognition result, the voting proportion of each group and a similarity average value. The technical scheme has the advantages of high recognition efficiency, accurate face recognition and high feature expression richness.

The invention discloses a tumor MRI weak supervised learning analysis modeling method, which comprises the following steps: S1, constructing a tumor MRI segmentation network based on a full convolutional neural network to realize tumor MRI preliminary segmentation; S2, taking the tumor MRI segmentation network as a generator, taking the classification model as a discriminator, and proposing a generative adversarial training algorithm of the tumor MRI segmentation network; and S3, taking unlabeled image data in the medical image as additional input condition information of the generator model and the discriminator model, guiding a data generation process, and enhancing the stability of the segmentation-generative adversarial network. The invention also discloses a model established by usingthe tumor MRI weak supervised learning analysis modeling method. According to the method, the tumor target region and the endangered organ can be more accurately and automatically segmented under thecondition of fewer training samples, and the method is applied to target region and endangered organ sketching of tumor radiotherapy.

The invention relates to a deep neural network target detection method integrated with a pixel-level attention mechanism. The deep neural network target detection method comprises the following steps:collecting a training sample image; converting data in the collected image data set into a format required by training the deep convolutional neural network through preprocessing, and performing dataenhancement to improve the training effect of the neural network; designing a deep convolutional neural network structure integrated with a pixel-level attention mechanism, the network structure being used for receiving an input image and outputting bounding box regression information and category information to an object in the image: the network structure comprising two parts, one part being abasic network for preliminarily extracting features; the other part being a plurality of cascaded convolutional layers or downsampling layers added at the tail end of the basic network and being usedfor extracting convolutional feature maps with different scales and fusing the multi-scale feature maps with a pixel-level attention mechanism; and training.

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 scheme relates to artificial intelligence, and provides an emotion recognition method, a device and a computer readable storage medium, andthe method comprises the following steps of: receiving audio data, generating a voice spectrum graph aiming at the audio data, inputting the voice spectrum graph to a first recognition module, inputting the audio data to a second recognition module, whereinthe first recognition module extracts a first feature vector by adopting a DCNN network and an RNN network which are sequentially connected the second recognition module extracts a MFCC from audio data, andthe MFCC is transformed into a second feature vector by s non-linear transformation;and the first and second feature vectors output from the first and second recognition modules are connected to form a combined feature and be input into a full connection layer and a soft Max layer in turn for emotional recognition.. By means of the scheme, the combination of the DCNN and the RNN is beneficial for simulating subtle local emotion clues, the MFCC describes voice characteristics in a short time, more emotion related features are captured through two features extraction modes, and the emotion recognition rate is improved.

The invention discloses a nasopharyngeal carcinoma auxiliary diagnosis model construction and auxiliary diagnosis method and system, and relates to the technical field of medical data processing. Thenasopharyngeal carcinoma auxiliary diagnosis model construction method comprises the steps: sample acquiring: acquiring a nasal endoscope image, wherein the nasal endoscope image comprises a nasopharyngeal carcinoma group and a non-nasopharyngeal carcinoma group; preprocessing: preprocessing the nasal endoscope image; and model training: inputting the preprocessed nasal endoscope image into a convolutional neural network, and training the convolutional neural network to obtain a nasopharyngeal carcinoma auxiliary diagnosis model. According to the invention, the nasopharyngeal endoscope image can be analyzed, and the predicted illness probability is output in real time to assist a doctor in nasopharyngeal carcinoma diagnosis, so that the accuracy of nasopharyngeal carcinoma diagnosis can beeffectively improved, and the biopsy detection rate is increased so as to achieve the purposes of early screening, early diagnosis and early treatment of nasopharyngeal carcinoma and improvement of treatment effect and prognosis of patients.

The invention discloses a sparse code division multiple access signal detection method based on a depth neural network. First, the relationship between users and resources is expressed as factor graph, Then the iterative transfer of information on the factor graph is transformed into the forward transfer of neural network, and then the signal detection error is estimated and the estimated value isused as the input data of the neural network, and then the gradient descent method is used to train the neural network to obtain better coefficients, and finally the trained network is used for SCMAsignal detection. Because of the use of neural network architecture, this method can be implemented on the corresponding AI chip. Compared with the traditional methods, this method can improve the performance and transfer the related computations to the high-speed parallel processing AI chip, which can effectively reduce the time delay caused by SCMA signal detection.

The invention discloses a method for counting targets in a remote sensing image based on an attention mechanism, which fuses the characteristics of the attention mechanism, a scale pyramid and deformable convolution on the basis of VGG16, and consists of three cascaded stages: feature extraction of a front-end network, scale fusion of a middle-end network and generation of a density map of a rear-end network. Through the technical scheme of the invention, the problems of target scale diversity, complex and disordered background interference and target direction randomness in a dense target counting task in the remote sensing image can be well solved.

The invention discloses a network user behavior prediction system, and the system comprises a data collection and storage module, a data preprocessing module, a user network behavior analysis module and a data presentation module, wherein the data collection and storage module, the data preprocessing module, the user network behavior analysis module and the data presentation module are connected sequentially. The data collection and storage module is used for collecting and storing the useful mobile Internet data of a user through collection equipment. The data preprocessing module is used for carrying out the data clearing and cleaning of the useful data, filtering the data comprising noise and abnormality, forming an effective data set of user's behavior analysis, and enabling the effective data set to be transmitted to the user network behavior analysis module. The user network behavior analysis module is used for carrying out the arrangement and analysis of the effective data set, carrying out the analysis of the behaviors of the user, and outputting a user's behavior analysis result. The data presentation module is used for presenting the user's behavior analysis result to the user. The system supports the analysis and mining of a large amount of mobile Internet data of the user, and is good in prediction effect.

The invention discloses a lesion area classification method and system for a full-view digital pathological section. The invention builds a CSResNet system, carries out learning training of the CSResNet system, achieves automatic segmentation of a lesion area in the full-view digital pathological section, classify a segmented lesion area, and judges the type of the lesion area. The residual attention module in the CSResNet system is combined with attention mechanisms in the channel direction and the space direction at the same time, so that the network can transfer the learning gravity centerto a key area, capable of deciding the category of an input image, in a feature map, and high calculation efficiency, classification precision and recognition capability are achieved.

The invention discloses a resource joint allocation method based on deep reinforcement learning in Internet of Vehicles. The method comprises the following steps: S1, constructing an Internet of Vehicles communication scene including vehicle-to-infrastructure communication and infrastructure-to-data center communication; S2, a base station collecting resource state information which can be allocated to vehicle nodes by communication cell infrastructure, and taking the resource state information as an input state of a deep reinforcement learning network DQN; S3, taking the connection state of the vehicle nodes and the infrastructure as an output action; S4, establishing an optimization model by taking maximization of the total throughput of vehicle node request tasks in a communication cell as a target; S5, designing a DQN reward function and a network structure, and training the DQN; and S6, according to the input state of the vehicle nodes, the DQN outputting an action with the maximum Q value as communication of the vehicle nodes, and calculating and caching a resource allocation strategy. According to the method, the joint allocation problem of communication, calculation and cache resources in the Internet of Vehicles is solved with lower complexity.