719 results about "Neural network learning" patented technology

Apparatus and methods for learning and training in neural network-based devices. In one implementation, the devices each comprise multiple spiking neurons, configured to process sensory input. In one approach, alternate heterosynaptic plasticity mechanisms are used to enhance learning and field diversity within the devices. The selection of alternate plasticity rules is based on recent post-synaptic activity of neighboring neurons. Apparatus and methods for simplifying training of the devices are also disclosed, including a computer-based application. A data representation of the neural network may be imaged and transferred to another computational environment, effectively copying the brain. Techniques and architectures for achieve this training, storing, and distributing these data representations are also disclosed.

An artificial intelligence system is provided which makes use of a dual subroutine to adapt weights. Elastic Fuzzy Logic ("ELF") System is provided in which classical neural network learning techniques are combined with fuzzy logic techniques in order to accomplish artificial intelligence tasks such as pattern recognition, expert cloning and trajectory control. The system may be implemented in a computer provided with multiplier means and storage means for storing a vector of weights to be used as multiplier factors in an apparatus for fuzzy control.

Time lapse, characteristic of aging, is a complex process that affects the reliability and security of biometric face recognition systems. Systems and methods use deep learning, in general, and convolutional neural networks (CNN), in particular, for automatic rather than hand-crafted feature extraction for robust face recognition across time lapse. A CNN architecture using the VGG-Face deep (neural network) learning produces highly discriminative and interoperable features that are robust to aging variations even across a mix of biometric datasets. The features extracted show high inter-class and low intra-class variability leading to low generalization errors on aging datasets using ensembles of subspace discriminant classifiers.

Determining semantically equivalent text or questions using hybrid representations based on neural network learning. Weighted bag-of-words and convolutional neural networks (CNN) based distributed vector representations of questions or text may be generated to compute the semantic similarity between questions or text. Weighted bag-of-words and CNN based distributed vector representations may be jointly used to compute the semantic similarity. A pair-wise ranking loss function trains neural network. In one embodiment, the parameters of the system are trained by minimizing a pair-wise ranking loss function over a training set using stochastic gradient descent (SGD).

Class types of input signals having unknown class types are automatically classified using a neural network. The neural network learns features associated with a plurality of different observed signals having respective different known class types. The neural network then recognizes features of the input signals having unknown class types that at least partially match at least some of the features associated with the plurality of different observed signals having respective different known class types. The neural network determines probabilities that each of the input signals has each of the known class types based on strengths of the matches between the recognized features of the input signals and the features associated with plurality of different observed signals. The neural network classifies each of the input signals as having one of the respective different known class types based on a highest determined probability.

The invention provides a fast magnetic resonance imaging method and device based on a deep convolution neural network. The fast magnetic resonance imaging method comprises steps of constructing a deep convolution neural network, obtaining an offline magnetic resonance image data, training the deep convolution neural network, learning a mapping relation between an undersampled magnetic resonance image and a full sampled image, using a deep convolution neural network learned through the S2 to reconstruct a resonance image. The fast magnetic resonance imaging method and device based on deep convolution neural network learn an offline deep convolution neural network through using substantial sampled magnetic resonance data, learns a mapping relation between the sampled magnetic resonance image and the full sampled image, fully uses offline substantial the magnetic resonance images to develop the prior information, enables the offline network to recover more fine structure and image characteristics from the undersampled resonance data and improves the magnetic resonance multiple and the imaging accuracy.

The invention discloses an on-chip learning neural network processor comprising a data interface module, a data preprocessing module, a data cache region module, a neuron weight cache region module, a random initialization module, a neural computing unit module, a neural network forward computing control module, an activation function module, a neural state controller module and a neural network learning algorithm control module. The neural state controller module controls all the unit modules to cooperatively work to perform neural network learning and reasoning. The neural computing unit module is designed by using general hardware acceleration computing for programmable control of the neural network computing type and computing scale. The assembly line technology is added in the design so that the data throughput and the computing speed can be greatly enhanced, the focus is put on optimization of the multiply-add units of the neural computing unit and the hardware area can be greatly reduced. Hardware mapping is performed on the neural network learning algorithm so that the neural network processor is enabled to perform on-chip learning and offline reasoning.

The invention relates to a detection method for codes plagiarism, which collects the similar values of two codes to be detected on different aspects as the input vector of an NN and obtains a conclusion whether plagiarism exists according to the study of the NN. The detection method can eliminate the effects of the codes on the two aspects of distribution and grammar equivalent transformation, can identify the plagiarism means on the aspect of semantic analysis as possible and improve the detection tool of the original detection tool. The detection method has a high detection precision and a good effect, can detect more plagiarism means and is suitable to be popularized.

The invention belongs to the technical field of computer vision and artificial intelligence, and particularly relates to a large-scale face recognition method based on a depth convolution neural network model. The method comprises steps of putting forward a residual error learning depth network model facing large-scale face recognition, wherein the residual error learning depth network model is formed by a convolution layer, a residual error layer and a full connection layer, and the residual error layer is formed by adding one path of multiple convolution layer cascade data and one path of original data to calculate the sum; and carrying out normalization operation in batch after each convolution layer in the model. According to the invention, by use of the characteristics of strong learning ability and good residual error learning convergence of the depth convolution neural network, layers of the model are increased in the aspect of the layer number of the network model; and in the aspect of residual error layer structure, the invention provides a highly efficient residual error layer structure. In the field facing the large-scale face recognition, the accuracy of the provided method is greatly improved compared with a base line model, and the accuracy of face retrieval in a million-class face recognition database can reach 74.25%.

The invention discloses a time sequence and CNN-based unsafe behavior identification method and system. The method comprises the steps of inputting original data of a training set video of unsafe behaviors to a CNN for performing training and learning, and outputting a spatial eigenvector from the last pooling layer of the CNN; taking the spatial eigenvector as an input of a time recurrent neuralnetwork, learning a time sequence dependency relationship contained in the spatial eigenvector by using the time recurrent neural network to learn a time sequence representation of spatial behaviors,and obtaining an average pooling eigenvector; inputting the average pooling eigenvector to a softmax layer, thereby obtaining a deep mixed learning model of the CNN and the time recurrent neural network, namely, a softmax classifier; and performing online identification on a monitoring video of a construction site by utilizing the softmax classifier to obtain the unsafe behaviors of the construction site. The support can be provided for real-time investigation and correction of the unsafe behaviors in a whole building engineering construction process.

The invention provides a signal identification and classification method. The method comprises the followings steps of: carrying out noise reduction on initial data containing higher noise by utilizing a wavelet transform method, decomposing signals into high-frequency information and low-frequency information in data analysis, carrying out noise cancelling on the signals by adopting a soft thresholding method and then carrying out signal reconstruction; carrying out further decomposition on the high-frequency part which is not detailedly classified by multiscale analysis while inheriting allthe favorable time-frequency localization advantages of the wavelet transform; analyzing the signals within different frequency bands after multi-layered decomposition by utilizing the wavelet packettransform to extract out characteristic information reflecting a system state; transforming the characteristic vectors of input signals into a high-dimensional characteristic space through non-lineartransform and then solving for an optimal linear classification plane in the high-dimensional characteristic space. The invention overcomes the defects of difficult determination of a network structure, low convergence rate, requirement on large quantities of data samples during training, and the like in neural network learning and enables the neural network learning to be with the characteristics of high precision and strong real time in the aspect of practical application of engineering.

The invention discloses a personalized recommendation system based on deep learning under a social network. The system mainly comprises an offline learning module and an online recommendation module. The offline learning module firstly generates a training sample seat to construct a deep convolutional neural network learning module with an attention mechanism and carries out iterative optimization on parameters in the learning module; and the online recommendation module carries out real-time item recommendation on a newly-registered user based on the learning model obtained through training. Compared with the prior art, the system has the advantages of high accuracy, fast speed and simplicity and easiness in implementation and can be effectively applied to the fields, such as electronic commerce, public opinion monitoring, intelligent transportation and medical treatment and health.

A method for configuring a neural network is provided. The method includes: selecting a neural network including a plurality of layers, each of the layers including a plurality of neurons for processing an input and providing an output; and, incorporating at least one switch configured to randomly select and disable at least a portion of the neurons in each layer. Another method in the computer program product is disclosed.

The invention discloses a risk evaluation method of an electric power communication network in the technical field of electric power communication. The risk evaluation method of the electric power communication network includes the steps of first collecting risk evaluation parameters of the electric power communication network, then structuring an index data base and a sample data base through the risk evaluation parameters, and finally training a neural network according to sample data in the sample data base and transferring the trained neural network to calculate corresponding risk value of the electric power communication network of index data in the index data base. According to the risk evaluation method of the electric power communication network, the index data is trained through the neural network, the inductive subjective factor in artificial given index weight is avoided, interference of odd data is avoided through network structure learning and network parameter learning, the quantity of redundancies in hidden layer nodes is reduced, neural network learning time is reduced, network learning speed is improved, corresponding index weight is adjusted automatically when novel risk factors appear, and thus risk evaluation method for the electric power communication network has good adaptivity and high precision.

The embodiment of the invention provides an indoor positioning method and device based on SLAM. The method comprises the steps of obtaining a to-be-positioned image uploaded by a user terminal, constructing a picture database with pose information rapidly through an SFM algorithm; extracting a feature vector of a to-be-positioned image and a feature vector set of a database image through a convolutional neural network; calculating the similarity between the feature vectors of the to-be-positioned image and the database image, completing loopback detection, and obtaining the image with the highest similarity with the to-be-positioned image in the image database and the first pose information corresponding to the image, thereby obtaining the accurate position and pose information of the userterminal. Compared with a traditional visual SLAM algorithm which adopts a word bag model and is weak in recognition capability, the deep features of the image are learned through the neural network,the higher recognition accuracy can be achieved, and the accuracy of loop detection is improved.

The invention relates to a robot walking control method based on the determined study theory. The method comprises the processes such as the establishment of robot walking model, the establishment of referenced gait model, the study on the neural network, the establishment of constant neural network, the completion of walking task by utilizing the constant RBF neural network, etc. The invention overcomes the disadvantages of learning ability in the current neural network study and the control method and realizes the accurate study of the unknown dynamic model of a robot closed-loop control system along the periodic gait track of the robot within a local area, and can study the effective knowledge of the system dynamics during the stable dynamic control process and apply the knowledge to the subsequent same or similar tasks successfully. The method not only can realize the rapid walking on the basis of energy saving, but also provides the powerful support for the development of the humanoid robots.

The invention provides a method for establishing a data analysis model, and the method comprises the steps: learning and pre-training an initial cloud model at a server side through employing a deep neural network based on the current user data, and transmitting the initial cloud model to different users; the user constructs a user model of the user based on the received cloud model according to the data characteristics of the user, learns and trains the user model by adopting a deep neural network, and returns the trained user model to the server; the server fuses the received user model based on a preset period to obtain a new cloud model; and the user adjusts the own user model based on the received new cloud model, learns and trains by adopting the deep neural network at the same time,and returns the trained user model to the server. According to the model framework constructed based on the method, the problems of data islands and individuation can be solved by combining federatedlearning and homomorphic encryption technologies, a powerful machine learning model is constructed by summarizing data from different organizations, and the privacy of the user can also be protected.

The invention discloses an oil reservoir inter-well connectivity determination method based on data driving, and the method comprises the steps of firstly collecting related parameters of an oil fieldblock, obtaining a filtering coefficient of each injection signal, carrying out the preprocessing of injection and production data, and correcting the time lag and attenuation of the injection and production data; performing normalization processing on the injection-production data to form a standard sample set of neural network learning and training; building a neural network, using a conjugategradient algorithm as a learning algorithm, and achieving the rapid optimization solution of neural network model parameters; and performing parameter sensitivity analysis based on the trained neuralnetwork model to obtain a connection coefficient for representing the inter-well connectivity of the oil reservoir. The method is simple and convenient, is high in calculation efficiency, is used forevaluating the dynamic connectivity between oil reservoir wells, has a better yield prediction effect while having the same inter-well connectivity coefficient calculation precision of a traditional inter-well connectivity judgment method, and can further guide formulation of optimization measures such as profile control and water plugging and intelligent oil field layered injection and productionhistory fitting and production optimization.

The invention provides an interest point recommendation method based on user positive and negative preference learning, and the method comprises the steps: learning deep features of interaction between a user and a scenic spot through employing a neural network, and training two neural network models: a positive preference neural model and a negative preference neural model; wherein the positive preference neural network model generates a list of scenic spots loved by the user, and then acquiring a final recommendation list through optimization of the negative preference neural network model,so that more accurate scenic spot recommendation is provided for the user. The problems that traditional interest point recommendation precision is not high, and the individuation degree of recommendation results is low are solved.

A defect inspection apparatus for inspecting an object to be inspected for a defect by processing an image taken from the object, includes: neural networks provided respectively for individual defect types to be classified; a learning unit which makes the neural networks learn based on the corresponding defect types to be classified; and a defect detection unit which classifies and detects defect types using the neural networks that have learned.

The invention discloses an agricultural machine path tracing controlled method based on nerve net, which comprises the following steps: adopting catalogue; exercising network; using network; finishing the method. This invention possesses self-adaptability, which can improve accuracy of track tracing effectively.

The invention discloses a deep neural network-based text consistency analysis method. After a section of text is input, a distributed method is adopted to translate each word in sentences into vectorsto form distributed sentence matrices, the words repeatedly appearing in the adjacent sentences are counted, and repeated information of the adjacent sentences is added through a manner of expandingdimensionality of the matrices; a convolutional neural network is utilized to learn distributed representations of the sentences, and important features of logic, semantics, syntax and the like in thesentences are extracted to form sentence vectors; and similarity degrees between the adjacent sentence vectors are calculated to add context association contents, finally, a neural network is continuously trained, and probability of text consistency is output. The method is characterized in that complex artificial feature extraction operations do not need to be carried out, external resources arealso not relied on, and compared with existing consistency analysis technology, the method provided by the invention has a great improvement in an accuracy rate, and has a better practical value.

The invention provides an unmanned aerial vehicle operator state evaluation method based on multi-sensor measurement and neural network learning. A mapping relation between a sensor characteristic and a semanteme is established, thereby estimating a decision grade of an operator. The method has advantages of facilitating more visual mastering of real-time operator state, estimating whether the operator is suitable for a task, and performing corresponding adjustment on the task. Because a neural network learning method is utilized, a characteristic space of different operators can be established. Therefore, the unmanned aerial vehicle operator state evaluation method is suitable for various grades of unmanned aerial vehicle operators with different proficiencies.