482 results about "Feedforward neural network" patented technology

A color management method/apparatus generates image color matching and International Color Consortium (ICC) color printer profiles using a reduced number of color patch measurements. Color printer characterization, and the generation of ICC profiles usually require a large number of measured data points or color patches and complex interpolation techniques. This invention provides an optimization method/apparatus for performing LAB to CMYK color space conversion, gamut mapping, and gray component replacement. A gamut trained network architecture performs LAB to CMYK color space conversion to generate a color profile lookup table for a color printer, or alternatively, to directly control the color printer in accordance with the a plurality of color patches that accurately. represent the gamut of the color printer. More specifically, a feed forward neural network is trained using an ANSI/IT-8 basic data set consisting of 182 data points or color patches, or using a lesser number of data points such as 150 or 101 data points when redundant data points within linear regions of the 182 data point set are removed. A 5-to-7 neuron neural network architecture is preferred to perform the LAB to CMYK color space conversion as the profile lookup table is built, or as the printer is directly controlled. For each CMYK signal, an ink optimization criteria is applied, to thereby control ink parameters such as the total quantity of ink in each CMYK ink printed pixel, and/or to control the total quantity of black ink in each CMYK ink printed pixel.

The invention belongs to the technical field of computers, and discloses a machine translation method and system based on a generative adversarial neural network. The method comprises the following steps that: on the basis of an original machine translation generation network, a discrimination network which generates network countermeasure with the original machine translation generation network is imported; a translation used for judging a target language is from a training parallel corpus and is a network machine translation result of the original machine translation generation network; and the discrimination network adopts a multi-layer sensor feedforward neural network model to realize binary classification. The system comprises the discrimination network, a generation network, a mono-lingual corpus and a parallel corpus. While manually annotated bilingual parallel corpus resources are fully utilized, and mono-lingual corpus resources also can be fully utilized to carry out semi-supervised learning; and the mono-lingual corpus resources are very rich and can be easily obtained, and the problem that required training corpora required by the neural network machine translation model are not sufficient is solved.

The invention discloses an aspect-level emotion classification model and method based on dual-memory attention, belonging to the technical field of text emotion classification. The model of the invention mainly comprises three modules: an encoder composed of a standard GRU loop neural network, a GRU loop neural network decoder introducing a feedforward neural network attention layer and a Softmaxclassifier. The model treats input statements as a sequence, based on the attention paid to the position of the aspect-level words in the sentence, Two memory modules are constructed from the originaltext sequence and the hidden layer state of the encoder respectively. The randomly initialized attention distribution is fine-tuned through the attention layer of the feedforward neural network to capture the important emotional features in the sentences, and the encoder-decoder classification model is established based on the learning ability of the GRU loop neural network to the sequence to achieve aspect-level affective classification capabilities. The invention can remarkably improve the robustness of the text emotion classification and improve the classification accuracy.

The invention discloses a feature loss-based medical image super-resolution reconstruction method. The method involves an image conversion network fw; the image conversion network fw is a feed-forward full connection neural network; neurons in the feed-forward full connection neural network are divided into different groups according to an information receiving sequence; each group is regarded as a network layer; the neurons in each layer receive numerical outputs of the neurons in the previous layer to serve as inputs of themselves, and outputs of themselves are input to the next layer; information in the whole network is transmitted in a direction; and the image conversion network fw receives a low-resolution medical image having a size of H/4XW/4 and sent by the feed-forward neural network, and converts the low-resolution medical image having the size of H/4XW/4 into a high-resolution image having a size of HXW.

Disclosed is a data analysis and cybersecurity method, which forms a time-based series of behavioral features, and analyzes the series of behavioral features for attack detection, new features derivation, and/or features evaluation. Analyzing the time based series of behavioral features may comprise using a Feed-Forward Neural Networks (FFNN) method, a Convolutional Neural Networks (CNN) method, a Recurrent Neural Networks (RNN) method, a Long Short-Term Memories (LSTMs) method, a principal Component Analysis (PCA) method, a Random Forest pipeline method, and/or an autoencoder method. In one embodiment, the behavioral features of the time-based series of behavioral features comprise human engineered features, and/or machined learned features, wherein the method may be used to learn new features from historic features.

The embodiment of the invention provides a regional air pollutant concentration prediction method, terminal and computer readable storage medium. The method includes using a calculated daily average historical pollutant concentration data set and a preprocessed daily historical meteorological data set as sample data sets, utilizing a random forest model to perform training, wherein the random forest model includes a plurality of decision trees, each decision tree being implemented by use of a multilayer feedforwad neural network; determining predicted meteorological data of a preset number ofdays in the future predicted on the same day at current time; preprocessing the predicted meteorological data; and according to the preprocessed predicted meteorological data and pollutant concentration data monitored on the same day at current time, utilizing the trained random forest model to predict pollutant concentration data of the preset number of days in the future of a region to be predicted. The regional air pollutant concentration prediction method provided by the embodiment of the invention improves regional air pollutant concentration prediction precision, and has relatively a high generalization capability.

The invention discloses a performance prediction method applicable to dynamic scheduling for a semiconductor production line. An extreme learning machine (ELM) is applied to prediction and modeling in the performance prediction method. Feeding control and scheduling rules are considered in a unified manner in the method, short-term scheduling key performance indexes such as an equipment utilization rate and a movement step number are predicted on the basis of a real-time state of a system, and a foundation is provided for dynamic real-time scheduling. A novel feed-forward neural network of the ELM is introduced into the semiconductor manufacturing system, and a prediction model is built by the aid of available data of the production line. As shown by test results, ideal prediction results can be quickly acquired by the method implemented by the aid of the ELM, the method has obvious advantages and an obvious application prospect in the aspects of parameter selection and learning speed as compared with the traditional neural network modeling method, and a new idea is provided for online optimal control.

The invention relates to an elevator risk evaluation method, and belongs to the technical field of elevator risk real-time evaluation and process and safety control. The elevator risk evaluation method comprises the steps that (1) an index analysis method and an analytic hierarchy process are used for analyzing factors which possibly cause elevator faults, and an elevator risk evaluation index hierarchical structure is built; (2) actually-measured parameter values of indexes of the elevator risk evaluation index hierarchical structure are obtained, and initialization processing is carried out on the actually-measured parameter values of the indexes; (3) the weight of the indexes of the elevator risk evaluation index hierarchical structure is determined through an artificial neural network; (4) risk evaluation is carried out on an elevator through a feed-forward neural network with a time sequence learning algorithm, and the elevator risk grades are divided. The elevator risk evaluation method has reference value and application value in a plurality of fields, and can be applied to risk evaluation of industrial boilers, large-sized power transformers, industrial pipelines and the like.

Embodiments of the present application disclose a named entity recognition method, a device, equipment, and a medium, wherein, the method includes: obtaining a text to be recognized; word segmentationprocessing being carried out on the text to be recognized to obtain a word segmentation sequence; inputting the word segmentation sequence to a named entity recognition model, and obtaining attributeidentifiers of named entities corresponding to each word segmentation output from the named entity recognition model; furthermore, the named entity in the text to be recognized being determined according to the attribute identification of the named entity corresponding to each participle. The named entity recognition model used in this method is based on feedforward neural network with simple network structure and fewer network parameters, which ensures that the model is easy to maintain and update. In addition, based on the multi-dimensional segmentation features that can fully and comprehensively express the semantic information of segmentation, the model determines the attribute identification of named entity corresponding to each segmentation, which ensures the accuracy of named entity recognition. In addition, the present application also provides a method and apparatus for training a named entity recognition model.

The invention discloses a feedforward neural network hardware realization method based on multicore technology, relating to a computer system based on a specific calculation model. The invention adopts an NIOS II soft core processor and utilizes Quartus II software in an FPGA development system thereof to build a multicore processor system, the number of processor cores is only limited by chip scale, each processor core can be independently programmed, and multiple processor cores can simultaneously run respective program and simulate the characteristics of nerve cells of any kind by performing programming of one single core; in the network studying process, the single cores calculate output amount in a parallel mode and finish the adjustment of corresponding weights and thresholds; the input and output of each single core are performed with data exchange via a multiport memory to stimulate feedforward neural network with any structure, which realizes the purposes of high speed, distribution and parallel calculation of the neural network.

The invention discloses a prediction energy management method of a networked hybrid electric vehicle. The prediction energy management method comprises the following steps that S1, a target vehicle uploads the own driving condition information to a data processing center through a vehicle-mounted terminal device; S2, the data processing center plans an optimal driving path of the target vehicle incombination with the collected road surface information and estimates a complete vehicle speed curve of the target vehicle; S3, the target vehicle receives the information feedback of the data processing center and sends the information feedback to a VCU for optimal energy distribution in combination with the real-time state information acquired by the target vehicle; S4, the VCU performs quick response planning on the received working condition based on a constructed two-layer feedforward neural network model to obtain a corresponding optimal global SoC trajectory; S5, the VCU follows the planned SoC trajectory through an MPC method, and obtaining an approximately optimal fuel economy energy distribution effect at a real-time control level. The method provided by the invention can ensurethat a real-time energy management strategy obtains the globally optimal fuel economy.

The present invention relates to a clinic pathology data classification method based on combination of principal component analysis and an extreme learning machine. The classification method comprises the following steps of (1) normalizing clinic data, extracting features through principal component analysis, ordering characteristic values according to characteristic significance, removing data dimensions below a significance threshold, and achieving the purpose of data dimension reduction; (2) training a feedforward neural network classifier by using an extreme learning algorithm after data dimension reduction is carried out; and (3) using the trained feedforward neural network classifier to test test samples, and obtaining a classification result. According to the clinic pathology data classification method based on combination of the principal component analysis and the extreme learning machine, dimensions are effectively reduced, calculation is simplified, and the classification effect is good.

The invention discloses a dynamic gesture learning and identifying method based on a Chebyshev neural network. Chebyshev orthogonal polynomials serve as hidden-layer neuron excitation functions for constructing a multi-input multi-output three-layer feedforward neural network, and a weights direct determination method and a hidden-layer node number adaptive determination algorithm are given; a fingertip detection algorithm based on a color histogram and a fingertip tracking algorithm based on bigraph optimal matching are given for obtaining a dynamic gesture track in real time; an MIMO-CNN (multi-input multi-output Chebyshev neural network) is subjected to input output structure design and network weights learning training according to the dynamic gesture identifying requirements, and a dynamic gesture is identified by the trained MIMO-CNN. A test result shows that the MIMO-CNN can increase the network training speed and improve the network training precision, so that the dynamic gesture learning speed is increased and the dynamic gesture identifying accuracy is improved; moreover, relatively good robustness and generalization ability in the aspect of dynamic gesture identification are achieved.

The invention discloses an image denoising method based on an ELM. The method comprises the following steps: establishing a basic feedforward neural network according to size of a to-be-processed image; aimed at the basic feedforward neural network, establishing a training sample set; using the training sample set to train the basic feedforward neural network based on an ELM method, to obtain a trained neural network; and inputting the to-be-processed image to the trained neural network, and the corresponding output being a de-noised image. Through customizing the training set and using the elm, the method trains connection parameters, so a training process can be completed rapidly, thereby greatly improving training efficiency. The method establishes the network training set according to noise types of an application scene, just the network training set established aimed at the application scene is needed to obtain the trained neural network aimed at different noise types to eliminate noise of an image. The method can be conveniently applied in different noise scenes.

The invention discloses a knowledge tracking method. The method comprises the following steps: converting a learning sequence and an exercise sequence into dense vector representation by utilizing a feedforward neural network, obtaining a correlation relationship and a relative position relationship of exercises by utilizing an attention mechanism and absolute positions of the exercises respectively, and initially summarizing knowledge states of students at each learning node of the learning sequence according to the two relationships; then, carrying out deep fusion on the preliminarily summarized historical knowledge states of the students by utilizing a convolutional neural network, and tracking continuously changing knowledge states of the students in a learning process; compared with the prior art, the problem of accumulated deviation in the knowledge tracking process can be reduced, the accuracy of knowledge tracking can be improved, and the efficiency of knowledge tracking can beeffectively improved; meanwhile, the method also greatly improves the robustness of a student knowledge state tracking process and the accuracy of a tracking result.

The invention discloses an industrial graph fusion method based on a graph convolutional neural network. The method is based on a plurality of constructed industrial sub-graphs, and the method comprises the following steps: constructing a local entity sub-graph of a graph; converting the structure embedding of the entity into the same vector space by using the attribute embedding of the attributetriple in the graph, forming an entity embedding vector, converting an entity alignment problem into a graph matching problem, and further forming a local matching vector by using a graph attention method; propagating local matching information in a graph through GCN to form a graph-level matching vector, and finally obtaining entity alignment in the graph through a double-layer feedforward neuralnetwork. According to the invention, the structural embedding of entities is converted into the same vector space through attribute embedding, the problem that pre-aligned entities are insufficient is solved, and the entity alignment problem in the graph is further converted into the graph matching problem through graph attention. Intelligence support is provided for optimizing the industrial structure, optimizing the regional structure and improving the industrial core competitiveness.

The invention provides a method for improving the voice quality of a throat microphone. A precise amplitude spectrum subjected to complete excitation influence removal is obtained by a STRAIGHT voice model, and a first resonance peak region and a second resonance peak region which play an important part in voice hearing perception of a throat in the amplitude spectrum are aggravated, so that a resonance peak weighing Mel cepstrum parameter and gain parameter pair as well as a line spectrum pair parameter and gain parameter pair which are suitable for voice conversion are obtained and respectively used as source and target characteristic references for conversion; and compared with the conventional cepstrum-cepstrum parameter pair, the line spectrum pair-line spectrum pair parameter pair and the Mel cepstrum-Mel cepstrum parameter pair, the obtained parameter pairs have a relatively high mapping relation. Furthermore, a designed dynamic feedforward neural network can automatically select a network topology structure; compared with a neural network with a fixed network structure, the dynamic feedforward neural network has relatively high generalization capacity and relatively high fitting precision; and therefore, the optimal nonlinear mapping function can be trained, and the voice quality is improved greatly.

The invention relates to a video content description method based on frame selection, comprising the following steps: a feed-forward neural network is used for constructing a screening model, and thescreening model screens the video frame according to the visual richness and semantic consistency of the video frame; constructing a description model for describing the content of the video to be described; training the screening model and the description model with training data; Selecting a description frame in the video to be described through the filtering model; the visual features of the description frame are extracted and the description model is input to obtain the description sentence of the video to be described.