796 results about "Neural net architecture" patented technology

A method useful for facilitating choosing a treatment or treatment regime and for predicting the outcome of a treatment for a disorder which is diagnosed and monitored by a physician or other appropriately trained and licensed professional, such as for example, a psychologist, based upon the symptoms experienced by a patient. Unipolar depression is an example of such a disorder, however the model may find use with other disorders and conditions wherein the patient response to treatment is variable. In the preferred embodiment, the method for predicting patient response includes the steps of performing at least one measurement of a symptom on a patient and measuring that symptom so as to derive a baseline patient profile, such as for example, determining the symptom profile with time; defining a set of a plurality of predictor variables which define the data of the baseline patient profile, wherein the set of predictor variables includes predictive symptoms and a set of treatment options; deriving a model that represents the relationship between patient response and the set of predictor variables; and utilizing the model to predict the response of said patient to a treatment. A neural net architecture is utilized to define a non-linear, second order model which is utilized to analyze the patient data and generate the predictive database from entered patient data.

Provided herein are methods and systems for implementing three-dimensional perception in an autonomous robotic system comprising an end-to-end neural network architecture that directly consumes large-scale raw sparse point cloud data and performs such tasks as object localization, boundary estimation, object classification, and segmentation of individual shapes or fused complete point cloud shapes.

The invention discloses a character detection method and device based on deep learning. The method comprises the steps: designing a multilayer convolution neural network structure, and enabling each character to serve as a class, thereby forming a multi-class classification problem; employing a counter propagation algorithm for the training of a convolution neural network, so as to recognize a single character; minimizing a target function of the network in a supervision manner, and obtaining a character recognition model; finally employing a front-end feature extracting layer for weight initialization, changing the node number of a last full-connection layer into two, enabling a network to become a two-class classification model, and employing character and non-character samples for training the network. Through the above steps, one character detection classifier can complete all operation. During testing, the full-connection layer is converted into a convolution layer. A given input image needs to be scanned through a multi-dimension sliding window, and a character probability graph is obtained. A final character region is obtained through non-maximum-value inhibition.

Fractal computers are neural network architectures that exploit the characteristics of fractal attractors to perform general computation. This disclosure explains neural network implementations for each of the critical components of computation: composition, minimalization, and recursion. It then describes the creation of fractal attractors within these implementations by means of selective amplification or inhibition of input signals, and it describes how to estimate critical parameters for each implementation by using results from studies of fractal percolation. These implementation provide standardizable implicit alternatives to traditional neural network designs. Consequently, fractal computers permit the exploitation of alternative technologies for computation based on dynamic systems with underlying fractal attractors.

A system and method for semantic extraction using a neural network architecture includes indexing each word in an input sentence into a dictionary and using these indices to map each word to a d-dimensional vector (the features of which are learned). Together with this, position information for a word of interest (the word to labeled) and a verb of interest (the verb that the semantic role is being predicted for) with respect to a given word are also used. These positions are integrated by employing a linear layer that is adapted to the input sentence. Several linear transformations and squashing functions are then applied to output class probabilities for semantic role labels. All the weights for the whole architecture are trained by backpropagation.

The invention discloses a multi-target tracking method based on a recurrent neural network, comprising the following steps: building a monitoring video data set marked with each frame of pedestrian location; manually expanding the monitoring video data set marked with each frame of pedestrian location to get training set samples; grouping the training set samples to get multiple training groups; building a multi-target tracking network; inputting the training groups in sequences to the multi-target tracking network for training; and inputting video data to be tested to the trained multi-target tracking network and performing forward propagation to get the moving trajectories of multiple targets. According to the invention, a proposed network model is trained in an end-to-end manner using raw data and a large amount of data obtained through artificial expansion, complex tasks such as data association and track estimation are completed under a unified neural network architecture, and the moving trajectories of targets can be tracked effectively under different directions, light conditions, deformations and other complex environments.

The invention relates to a target detection algorithm based on multi-feature extraction and multitask fusion. The technical characteristics of the target detection algorithm are that image features are extracted based on a deep learning convolutional neural network framework, the multilayer convolutional output result is extracted to form a multi-feature graph, and target areas-of-interest of different horizons are extracted from the multi-feature graph and feature connection is performed; semantic segmentation of the original graph is performed and the target segmentation area result is extracted, and multitask cross auxiliary target detection is performed on the target detection result and the target segmentation result in full connection layers through certain proportionality coefficient; and the result passes through the last full connection layer and then the image features are classified and regressively positioned through the combination classification and positioning loss function so that the final target detection result can be obtained. High-precision target detection positioning and classification can be realized by feature extraction through the deep learning convolutional neural network, multi-group and multilayer fusion and connection of the image features and loss function combination so that the great target detection result can be obtained.

The invention discloses a commodity target word oriented emotional tendency analysis method, which belongs to the field of the analysis processing of online shopping commodity reviews. The method comprises the following four steps that: 1: corpus preprocessing: carrying out word segmentation on a dataset, and converting a category label into a vector form according to a category number; 2: word vector training: training review data subjected to the word segmentation through a CBOW (Continuous Bag-of-Words Model) to obtain a word vector; 3: adopting a neural network structure, and using an LSTM(Long Short Term Memory) network model structure to enable the network to pay attention to whole-sentence contents; and 4: review sentence emotion classification: taking the output of the neural network as the input of a Softmax function to obtain a final result. By use of the method, semantic description in a semantic space is more accurate, the data is trained through the neural network so as to optimize the weight and the offset parameter in the neural network, parameters trained after continuous iteration make a loss value minimum, at the time, the trained parameters are used for traininga test set, and therefore, higher accuracy can be obtained.

The invention relates to a deep-learning-based classification and grading method for diabetes retinopathy. For each kind of diabetes retinopathy, lots of ophthalmoscope photographs are prepared; a deep convolution neural network including multi-level neural network architecture is established; the deep convolution neural network is trained based on the lots of ophthalmoscope photographs, so that the final output value of the deep convolution neural network meets a grading result of the ophthalmoscope photographs; and thus disease grading can be carried out by using the trained deep convolution neural network. According to the method provided by the invention, with application of the lots of ophthalmoscope photographs including diagnosis marks, needed features are learned automatically from a training example base by deep learning and grading diagnosis is carried out; and data features for determination and the deep convolution neural network parameters are corrected continuously during the training process, so that the grading accuracy and reliability in a real application scene can be improved substantially.

The invention discloses a BP neural network construction and identification method for identifying different brown rice grains, adopting the following technical scheme: including the following steps: 1) acquiring images: 2) image preprocessing: 3) extracting image feature information of different brown rice grains: 4) main Component analysis reduces dimensionality of image feature information; 5) Design BP neural network structure: 6) Train neural network, and use the BP network neural network constructed by any of the above-mentioned construction methods to identify different brown rice grains. The method proposed by the present invention obtains images of different types of brown rice grains through black as the background, uses image processing technology to obtain its characteristic information, and uses principal component analysis to reduce the dimensionality of the characteristic information, and finally uses BP neural network to perform different types of brown rice grains. identify. The method can identify different types of grains objectively, accurately and quickly, and overcomes the disadvantages of traditional manual detection.

The invention discloses a recommendation method based on deep social relationship, which comprises the steps of 1, constructing a rating matrix of users for an object and a social relationship matrixamong the users; 2, constructing an input layer in a one-hot coding mode; 3, obtaining a social embedded matrix by automatic coding according to the social relationship matrix among the users; 4, constructing an embedded layer through the input layer and the social embedded matrix; 5, constructing a collaboration layer through the embedded layer; and 6, inputting information of the collaboration layer into a hidden layer to obtain corresponding predicted score information, and thus performing object recommendation. According to the invention, the social relationship among the users can be fully utilized to solve a problem of data sparsity, excellent accuracy is ensured at the same time through establishing a complex deep neural network structure between users and objects, and thus accuraterecommendation is carried out for the users.

A deep approximation neural network architecture which extrapolates data over unseen states for demand response applications in order to control distribution systems like product distribution systems of which energy distribution systems, e.g. heat or electrical power distribution, are one example. The method is a model-free control technique mainly in the form of Reinforcement Learning (RL) where a controller learns from interaction with the system to be controlled to control product distributions of which energy distribution systems, e.g. heat or electrical power distribution, are one example.

The invention relates to a generative adversarial network method fusing a self-attention mechanism, belongs to the field of computer vision, and particularly relates to a generative adversarial network for image generation. The generation of the image is an important challenge in the field of computer vision, and if a large number of high-quality image samples can be generated, the artificial intelligence field can be more rapidly developed in the era depending on the big data background. Therefore, the invention provides the generative adversarial network fusing the self-attention mechanism,the network can generate the high-quality images, and meanwhile, the images have higher diversity. Specifically, the generative adversarial network uses a Waserstein distance to measure an evaluationcriterion of the generator and discriminator distribution, and a loss function is correspondingly improved; meanwhile, the self-attention mechanism is introduced into the neural network architecture corresponding to the generator and the discriminator, so that the relevance between the local pixel regions of the generated image is improved, and the quality of the generated image is improved.

The invention discloses a target tracking method based on an LSTM neural network, and belongs to the technical field of target tracking. According to the method, the LSTM is used for tracking the complex and non-linear motion target, and the problems that target tracking is difficult, a target model is difficult to establish and the tracking precision is low are solved; the method comprises the following steps: firstly, acquiring latitude and longitude information and speed information of a target, and processing acquired data; designing an LSTM neural network structure for single target tracking; and finally, adjusting LSTM neural network parameters to realize target tracking. According to the method, the nonlinear filtering process is effectively simplified, and a complex nonlinear target can be effectively tracked; the establishment of a target motion model and the utilization of a traditional filtering algorithm are not needed; estimating the target motion state of the next momentby using historical target motion information; adjusting internal parameters of the neural network by using a back propagation algorithm; the learning rate attenuation method reduces the calculation amount and improves the precision.

Systems, methods and computer-readable medium for (i) accelerating the inference speed of a deep neural network (DNN), and (ii) compressing the vector representations produced by the DNN out of a variety of input data, such as image, audio, video and text. A method embodiment takes as inputs a neural network architecture and a task-dependent loss function, measuring how well a neural network performs on a training data set, and outputs a deep neural network with sparse neuron activations. The invented procedure augments an existing training objective function of a DNN with regularization terms that encourage sparse activation of neurons, and compresses the DNN by solving the optimization problem with a variety of algorithms. The present disclosure also shows how to utilize the sparsity of activations during the inference of DNNs so the number of arithmetic operations can be reduced proportionately, and how to use the sparse representations produced by the DNNs to build an efficient search engine.

The invention provides a video driver behavior identification method based on a multi-task space-time convolutional neural network. A multi-task learning strategy is introduced into a training processof a space-time convolutional neural network and is applied to driver behavior identification in a monitoring video. Auxiliary driver positioning and optical flow estimation tasks are implicitly embedded into the video classification tasks, so that the convolutional neural network model is promoted to learn richer driver local space and motion time characteristics, and the driver behavior recognition accuracy is improved. Compared with an existing driver identification method, the multitask space-time convolutional neural network architecture designed by the invention combines interframe information, is high in generalization and identification accuracy, can be used for real-time driver behavior identification under a monitoring video, and has an important application value in the field of traffic safety.

The invention provides a wind power plant cluster short-term power prediction method based on a space-time diagram convolutional neural network, and the method comprises the steps: collecting historical power in a first target time period to obtain a historical power vector time sequence, and collecting weather forecast parameter vectors in a second target time period to obtain a weather forecastparameter matrix time sequence; inputting the historical power vector and the time sequence of the weather forecast parameter matrix into a prediction model, and outputting a predicted power vector time sequence in a third target time period, wherein the prediction model is obtained by training based on a sample historical power vector, a sample weather forecast parameter matrix timing sequence and a prediction power vector timing sequence label, and a neural network structure of the prediction model is formed based on a Bi-GRU network and a graph convolution network. According to the method provided by the invention, two factors of historical power and weather forecast parameters can be jointly considered in power prediction, and the prediction accuracy is also improved.

The invention provides an audio data generating method for voice synthesis. The audio data generating method comprises the following steps: extracting the text features in the text data to obtain textfeature data; carrying out the acceleration conversion processing on the text feature data through a neural network structure, and converting the text feature data into acoustic feature data; performing sound synthesis or selected splicing according to the acoustic feature data to obtain audio data. According to the invention, through the special anti-convolution structure is adopted, a good voice synthesis effect can be achieved on the premise that no any auto-regressive structure is included and few parameters are used, the calculation delay can be reduced while the prediction precision ofthe acoustic features can be guaranteed through the neural network structure, and the requirement for computing resources is reduced, the concurrent quantity is increased, the voice synthesis speed isincreased, and contribution is made to improvement of human-computer interaction experience.