629 results about "Perceptron" patented technology

The invention relates to the technical field of traffic control and artificial intelligence and provides a traffic signal self-adaptive control method based on deep reinforcement learning. The method includes the following steps that 1, a traffic signal control agent, a state space S, a motion space A and a return function r are defined; 2, a deep neutral network is pre-trained; 3, the neutral network is trained through a deep reinforcement learning method; 4, traffic signal control is carried out according to the trained deep neutral network. By preprocessing traffic data acquired by magnetic induction, video, an RFID, vehicle internet and the like, low-layer expression of the traffic state containing vehicle position information is obtained; then the traffic state is perceived through a multilayer perceptron of deep learning, and high-layer abstract features of the current traffic state are obtained; on the basis, a proper timing plan is selected according to the high-layer abstract features of the current traffic state through the decision making capacity of reinforcement learning, self-adaptive control of traffic signals is achieved, the vehicle travel time is shortened accordingly, and safe, smooth, orderly and efficient operation of traffic is guaranteed.

The invention discloses a collaborative filtering recommendation method and a collaborative filtering recommendation system of a convolution neural network integrating an attention model, which relates to the technical field of data mining recommendation, improves feature extraction efficiency and scoring prediction accuracy, reduces operation and maintenance cost, simplifies cost management mode,and is convenient for joint operation and large-scale promotion and application. The invention relates to a collaborative filtering recommendation method of a convolution neural network integrating an attention model, comprising the following steps: step S1, splicing a user feature vector and an item feature vector into a new vector; S2, sending the new vector as an input vector into the multi-layer perceptron to learn and predict the score; The attention model is fused into the object potential vector to obtain the convolution neural network of the object feature vector or the hidden layer of the multi-layer perceptron.

An approach for phoneme recognition is described. A sequence of intermediate output posterior vectors is generated from an input sequence of cepstral features using a first layer perceptron. The intermediate output posterior vectors are then downsampled to form a reduced input set of intermediate posterior vectors for a second layer perceptron. A sequence of final posterior vectors is generated from the reduced input set of intermediate posterior vectors using the second layer perceptron. Then the final posterior vectors are decoded to determine an output recognized phoneme sequence representative of the input sequence of cepstral features.

The invention discloses a three-dimensional target detection method, apparatus and device based on multi-sensor information fusion, and a computer readable storage medium. The three-dimensional targetdetection method comprises the steps: fusing 3D point cloud and an RGB image collected by a laser radar and a camera sensor, and generating an RGB-I image; generating a multi-channel aerial view according to the 3D point cloud so as to determine a region of interest; respectively extracting and fusing region-of-interest features of the RGB-I image and the aerial view based on a convolutional neural network; utilizing a multi-layer perceptron to fuse the confidence coefficient, the approximate position and the size of the image prediction target based on the features of the region of interest,and determining a candidate box; adaptively endowing different pixel weights to different sensor candidate box feature maps based on an attention mechanism, and carrying out skip fusion; and processing the candidate frame feature fusion image by using a multi-layer perceptron, and outputting a three-dimensional detection result. According to the three-dimensional target detection method, apparatus and device, and the computer readable storage medium provided by the invention, the target recognition rate is improved, and the target can be accurately positioned.

The invention discloses a deep convolutional neural network-based human face occlusion detection method. The method comprises the steps of performing block segmentation on an input image to obtain a target pre-selected region; constructing a first deep convolutional neural network, training the first deep convolutional neural network comprising a first deep convolutional network and a first multilayer perceptron connected with the first deep convolutional neural network to obtain required parameters, extracting features of the target pre-selected region, and performing classification; predicting the position of a human head through a second multilayer perceptron according to the extracted features; filtering the credibility of a classification type which is the human head and the predicted position of the human head through non-maximum suppression to remove an overlapped duplicate detection box; and obtaining a human head block in combination with original image segmentation, constructing a multi-task learning policy-based second deep convolutional neural network, and judging whether the left eye, the right eye, the nose and the mouth of the human head block are occluded or not. According to the method, the occluded human face can be accurately detected and the specific occluded part of the human face can be judged; and the method is mainly used for crime pre-warning of videos of a camera in front of an automatic teller machine.

The invention belongs to the technical field of cloud computing and discloses an efficient and privacy-preserving single-layer perceptron learning scheme in a cloud computing environment. The scheme comprises the steps that a client provides a security parameter, operates a key generation algorithm of a symmetric homomorphic encryption algorithm to calculate a public parameter and a key, then operates an encryption algorithm, encrypts training data through utilization of the key to obtain a corresponding ciphertext, and sends the ciphertext and related expectation to a cloud server, assists acloud server to judge a positive or negative characteristic of a dot product result in a training process, and decrypts the ciphertext of the received final optimum weight vector after a training taskis finished, thereby obtaining a single-layer perceptron prediction model; and the cloud server stores the training mode, trains a single-layer perceptron model and sends the ciphertext of the finaloptimum weight vector to the client after the training task is finished. The safety analysis shows that according to the scheme, in the training process, the privacy of the training data, an intermediate result and the optimum prediction model can be preserved, and the scheme is efficient in computing overhead and communication overhead aspects.

A database has a plurality of entries and a plurality of attributes common to each entry, where each entry corresponds to an individual. A query is received from a querying entity query and is passed to the database, and an answer is received in response. An amount of noise is generated and added to the answer to result in an obscured answer, and the obscured answer is returned to the querying entity. The noise is normally distributed around zero with a particular variance. The variance R may be determined in accordance with R>8 T log2(T / δ) / ε2, where T is the permitted number of queries T, δ is the utter failure probability, and ε is the largest admissible increase in confidence. Thus, a level of protection of privacy is provided to each individual represented within the database. Example noise generation techniques, systems, and methods may be used for privacy preservation in such areas as k means, principal component analysis, statistical query learning models, and perceptron algorithms.

A multithreaded microprocessor includes an instruction fetch unit including a perceptron-based conditional branch prediction unit configured to provide, for each of one or more concurrently executing threads, a direction branch prediction. The conditional branch prediction unit includes a plurality of storages each including a plurality of entries. Each entry may be configured to store one or more prediction values. Each prediction value of a given storage may correspond to at least one conditional branch instruction in a cache line. The conditional branch prediction unit may generate a separate index value for accessing each storage by generating a first index value for accessing a first storage by combining one or more portions of a received instruction fetch address, and generating each other index value for accessing the other storages by combining the first index value with a different portion of direction branch history information.

The invention discloses a living body detection method, a computer device and a computer readable storage medium. In the living body detection method, a multi-layer perceptron is trained by using a preset training set to determine a multi-layer perceptron model, continuous N-frame human face images to be detected is obtained, an intermediate frame human face image of the consecutive N-frame humanface images is switched from a first color space to a second color space, textural features of the intermediate frame human face image and dynamic pattern features of the continuous N-frame human faceimages are extracted, fusion features are acquired by fusing the textural features and the dynamic pattern features, the multi-layer perceptron model is used to perform feature mapping on the fusionfeatures, mapping features are output and normalized, a predicted probability value of a living body tag and a predicted probability value of a non-living body tag are obtained, and then the continuous N-frame human face images are determined as living body or non-living body human face images. The fusion features include the textural features and the dynamic pattern features, and thus the recognition accuracy and safety of living body detection can be improved.

The invention is an add-on implementation of a stabilized association memory matrix system to an existing convolutional neural network framework. This invention emulates the intra-action and the inter-action of the cognitive processes of the (logical) left-brain and (intuitive) right-brain. The invention is a numerically stable soft-ware based implementation that (1) reduces the long training times, (2) reduces the execution time, and (3) produces intralayer and interlayer connections. The implementation of this joint processing architecture is designed to take an existing hierarchy of stepped based processes, add next to it a parallel hierarchy of associative memory processes, and then connect the two processes by another set of associative memory processes. Or, the stepped-based process may be replaced with additional associative memory processes to enhance the emulation of several bidirectional intralayer and interlayer cognitive process communication. In addition, the invention can be used as a neural network layer compression tool that takes in a multilayer perceptron, also known as a multilayer neural network, and outputs a single layer perceptron. The final construction can be visualized as two vertical rails connected with a set of horizontal rungs which motivates the name to this invention: J. Patrick's Ladder: A Machine Learning Enhancement Tool.

The invention relates to the image processing area including following steps. (1) With being read from the USB port by the computer, the target images to be processed are stored in the memory. (2) The images to be compressed are carried out the lossless color sphace transformation. (3) The current pixels are predicted by using the two dimensions weighted predication model. The predicted residual is calculated. (4) Whether the mapped predicted residual is less than the preset error limit or not is determined. The self-adopting adjustment is needed if the residual exceeds the threshold. (5) The RICE entroy coding is accepted. (6) The compressed result of the image is output and saved. The invented method possesses the advantages of lower complexity, higher compression ratio and high executing speed.

A machine-aided smell identification method based on modularized combination neural net characterized by that, the combination neural net categorizer comprises a forward direction recessive single layer perceptron module and a RBF neural net module, wherein each categorizer module includes a plurality of sub-modules, whose arrangement is determined through the two phases of accretion and trimming, the parameters of center, width and weight value can be determined through error reversion algorithm. The invention solves the problem of the high dimension multiple catalog mass sample set in a fast and effective way. The machine-aided device by the invention can be used to distinguish many kinds of smells and to estimate the odor intensity simultaneously.

The invention discloses an intelligent diamond cutter with a real-time sensing and monitoring system and a cutter body matched with the intelligent diamond cutter, and belongs to the field of ultra-precision cutting machining and cutting state real-time monitoring. The requirements of real-time sensing and monitoring and adaptive machining of an ultra-precision machining process can be met. During the cutting machining of the intelligent diamond cutter, the cutting temperature, cutting force and vibration of the cutter are measured through a micro-electromechanical sensing system. The diamondcutter is fixedly connected to the front end of a cutter substrate. The micro-electromechanical sensing system is arranged in a cavity in the cutter substrate. A sensing and measurement long arm overhanging beam extended into an acute-angled cutter point area is arranged on the micro-electromechanical sensing system. A temperature sensor is arranged on the end part of the sensing and measurement long arm overhanging beam. Strain sensors are symmetrically arranged on the upper and lower surfaces of the root of the sensing and measurement long arm overhanging beam. An acceleration sensor and aninformation processing and wireless transmission module are integrated in the micro-electromechanical sensing system. A cavity is sealed by an upper sealing cover and a rear sealing gasket. The intelligent diamond cutter is arranged on a matched shank. The intelligent diamond cutter and the cutter body are used for the ultra-precision cutting machining and the real-time monitoring of the cutting machining process.

The invention provides a Chinese lexical analysis method based on a linear model, comprising the following steps: 1) a Chinese sentence is input and the length of an analysis window is set; 2) the verbatim analysis is carried out to the sentence, the character or a character set of each character in the sentence in a time window is input in a perceptron classifier, thus obtaining the score of a perceptron model which tags the current character as the certain word segmentation tag and part-of-speech tag; at the same time, the character or the character set of the character in the time window is input in the linear lexical analysis model, thus obtaining the score of the linear lexical analysis model which tags the current character as the certain word segmentation tag and the part-of-speech tag; 3) the score of the perceptron model and the score of the linear lexical analysis model are carried out the weighted sum, thus obtaining the comprehensive analysis score, the word segmentation tag and the part-of-speech tag with the highest comprehensive analysis score are taken as the word segmentation tag and the part-of-speech tag of the current character; when the word segmentation tag and the part-of-speech tag of all the characters complete the tagging, the lexical analysis of the Chinese sentence is completed. The Chinese lexical analysis method can significantly improve the accuracy of segmentation and tagging.

The invention belongs to the field of natural language processing text emotion analysis, and relates to a method for identifying a text emotion reason identification method. The method mainly comprises the following steps: obtaining a text containing candidate emotion reasons and emotion descriptions; converting the clauses into a word embedding matrix; usingbi-directional long short term memory network Bi-LSTM to encode the context information of the clauses; learning semantic relationships between the emotion description clauses and the candidate cause clauses by using an attention mechanism; for the emotion description clause set, extracting a local maximum semantic meaning by using a convolutional neural network CNN; using the Bi-LSTM to encode context information between the clauses;splicing the emotion description clause set and the coded candidate reason clause, and judging whether the emotion description clause set and the coded candidate reason clause have an emotion initiation relationship or not by using a multi-layer perceptron network MLP. According to the method, the problem that semantic relations between the emotion reason clauses and the emotion description clauses and between the emotion reason clauses are not fully considered in a traditional method is solved. Therefore, the invention provides a method for fusing the context of the clause and the context ofthe sentence, so that the emotion reason identification accuracy is improved.

The invention relates to a coarse-grained emotion analysis method based on a hierarchical BERT neural network, and belongs to the technical field of Web mining and intelligent information processing.The method comprises the following steps: corpus acquisition: acquiring a corpus of coarse-grained sentiment analysis; corpus preprocessing, wherein character cleaning, subordinate clause segmentationand subordinate clause vector construction are included; constructing sentence vectors: calculating the subordinate clause vectors by utilizing a bidirectional long and short term memory network, a multi-layer perceptron and an attention mechanism to generate sentence vectors; gradient coordination mechanism optimizing: introducing the gradient coordination mechanism to solve the problem of datatype imbalance in coarse-grained sentiment analysis; and carrying out coarse-grained sentiment analysis by adopting a hierarchical BERT neural network. Compared with the prior art, t the sentence vectors containing deep semantic information are constructed for the comment text through the hierarchical BERT neural network, the accuracy of coarse-grained emotion analysis tasks is improved, and the method has a wide application prospect in the fields of information recommendation, public opinion monitoring and the like.

A method for calculating a neuron layer of a multi-layer perceptron model that includes a permanently hardwired processor core configured in hardware for calculating a permanently predefined processing algorithm in coupled functional blocks, a neuron of a neuron layer of the perceptron model being calculated with the aid of an activation function, the activation function corresponding to a simplified sigmoid function and to a simplified tank function, the activation function being formed by zero-point mirroring of the negative definition range of the exponential function.

The invention discloses an intelligent auxiliary diagnosis method based on deep learning and ensemble classification. Through named entity identification and relation extraction, entities and attributes in the chief complaint and the current medical history are accurately extracted, and invalid information is removed. In the label topic model, the position weight of the feature words is added, andthe proportion of the feature words at the key position is increased. Adjusting parameters are added into a loss function of a multilayer perceptron model, so that the problem caused by uneven sampledistribution is solved. For the same sample, the classification boundaries obtained by different classification methods are different, so that the label topic model and the multi-layer perceptron model are integrated by adopting a stacking integration method, and the disease prediction accuracy is improved.

A received signal strength (RSS) predictor of a propagation-free model is trained based on a large number of actual network data sets, and the coverage performance of base station (BS) deployment is optimized through a multi-target heuristic method. In particular, more practical features of signal propagation, such as geographical environment and operating parameters of a base station, are fed into a machine learning (ML) model to predict received signal strength; besides, a multi-objective greedy algorithm is designed based on the prediction model, a feasible solution is initialized to meet geographical constraints and is fixed to be related to longitude and latitude of an optimization area, the optimization step length in the search direction is fixed, the step length is set according tothe upper limit and the lower limit of parameters, and the optimization objective is that the coverage rate reaches the standard with the least base stations. Numerical simulation results show that the multi-layer perceptron is superior to other machine learning algorithms in the aspect of received signal strength prediction, convergence and availability of the method are verified through base station deployment simulation, the coverage rate is better than that of actual deployment, and the number of base stations needing to be deployed is smaller.

The invention relates to the field of computer vision and deep learning, in particular to a video group behavior recognition method based on cascade Transformer, which comprises the following steps: firstly, acquiring and generating a video data set, extracting three-dimensional spatial-temporal features from the video data set through a three-dimensional backbone network, and selecting a key frame image spatial feature map; preprocessing the key frame image spatial feature map, sending the preprocessed key frame image spatial feature map into a human body target detection Transformer, and outputting a human body target box in the key frame image; then, mapping a sub-feature map corresponding to the screened human body target box on the key frame image feature map, calculating query / key / value in combination with a key frame image surrounding frame feature map, inputting the query / key / value into a group behavior recognition Transfomer, and outputting a group level spatial-temporal coding feature map; and finally, classifying group behaviors through a multi-layer perceptron. The method has the effect of effectively improving the group behavior recognition accuracy.

The invention discloses a personalized recommendation method and system based on interactive data clustering. The method comprises the following steps: constructing a user project interaction characteristic matrix, constructing a user historical behavior matrix, classifying the interaction matrix P by using a classifier, clustering similar users, selecting adjacent projects, training a neural network, and carrying out personalized recommendation. According to the invention, historical information such as browsing records and user search records of a user to a project and user personal information are spliced to construct user characteristics; the user characteristics are clustered by using a clustering algorithm, and personalized recommendation is performed on specific users on the basis,so that the performance of a traditional matrix decomposition model is improved, matrix decomposition is combined with a multi-layer perceptron, and the relationship between user projects is learned and predicted. And the recommendation precision in a big data environment is improved.

The invention discloses a text classification method and system based on label information and text characteristics and a medium, and the method comprises the steps: a characteristic extraction step:extracting a first text characteristic from a given text, and extracting a first label characteristic from a label corresponding to the given text; a feature fusion step: performing feature fusion onthe first text feature and the first tag feature to obtain a fused third feature; a model training step: inputting the fused third features into a multi-layer perceptron to train the multi-layer perceptron so as to obtain a trained multi-layer perceptron; text classification step, extracting a second text feature from the to-be-classified text, and extracting a second label feature from the labelcorresponding to the text to be classified, carrying out feature fusion on the second text feature and the second label feature to obtain a fused fourth feature, inputting the fused fourth feature into the trained multilayer perceptron for classification, and outputting a classification result.

The invention relates to a sentence alignment method based on a depth neural network, which comprises the following steps: preprocessing corpus, generating a word list and a word embedding word list,adopting a bi-directional circulating neural network layer to code the sentence, not only considering the semantic information of the word itself, but also considering the context information of the word, so that each word obtains a hidden state containing the context information of the word; The hidden states of words in each sentence are averaged to obtain sentence vectors, and then the two sentence vectors are stitched together. The perceptron layer is used to obtain a more abstract representation to determine whether the sentences are aligned or not. In addition, the word concealment stateobtained by the bidirectional loop neural network encoding can not only contain its own meaning, but also contain its context information.

The invention discloses a knowledge map relation inference algorithm based on a stacked neural network, and belongs to the technical field of artificial intelligence representation learning. The invention constructs a stacked neural network model comprising two components: a standard LSTM loop neural network and a multi-layer perceptron network. The model regards the triplets in the knowledge mapas short sentences and uses the learning ability of LSTM loop neural network to model the logical and semantic characteristics of the knowledge map in order to learn the grammatical and semantic information in the knowledge map. Through the bottom-level feature learning process, the upper-level fully connected network can provide enough discrimination to distinguish objects in different contexts.The algorithm of the invention models the three tuples in the knowledge map from the perspective of semantics, fundamentally solves the semantic diversity of entities and relationships, can greatly reduce the computational overhead of the relational reasoning of the large-scale knowledge map, and simultaneously can ensure the effectiveness of the relational reasoning.

The invention discloses an air pollutant concentration prediction method based on a graph attention mechanism. The method comprises steps of constructing a spatial pollutant concentration prediction model based on a graph attention mechanism; and taking the meteorological data, the air monitoring data and the environmental factor data as model input data, constructing a graph adjacency matrix through a graph attention mechanism, extracting graph information characteristics by utilizing a graph convolutional neural network layer and a multi-layer perceptron network layer, and outputting a predicted air pollutant concentration value. According to the method, the air pollutant concentration prediction is more accurate, and the process is more efficient.

The invention discloses a sensor residual self-coding network seismic data denoising method, and aims to overcome the defect that a traditional denoising method cannot meet the high precision exploration requirement on the aspects of generalization ability, denoising fidelity and the like. Characteristic data coded and reconstructed by a convolution self-coding network are output by a decoding part after passing through a multilayer sensor residual convolution self-coding block constructed by a multi-layer sensor convolution layer, a multi-scale layer, a BN layer and an auto-encoder, and combined with an optimized convolution core, so that the sensor residual self-coding network seismic data denoising method based on deep learning is put forward. The seismic data denoising method disclosedby the invention can still remove multiple waves and random noise at the same time while completely retaining the local details of seismic data and generating no false impression.

The invention discloses a text event extraction method in combination of sparse coding and a structural perceptron. The method comprises following steps: 1) normatively labeling and creating text data according to ACE or RichERE as training samples; 2) taking entities extracted as candidate entities for event trigger words and event parameters and extracting text features; 3) further extracting text distributive word vector features and learning sparse coding features; 4) utilizing training samples and extracted text features, training a classifier of the structural perceptron while recognizing trigger words and parameters related to events in texts; 5) inputting the classifier of the structural perceptron through the step 1 as for new text data and extracting text event information. The text event extraction method in combination of sparse coding and the structural perceptron has following beneficial effects: sparse coding expressions for distributive word vector features based on a neural network are utilized for enhancing text features; on the other hand, a model of the structural perceptron is utilized for learning event trigger words and recognizing event participants. Therefore, the better event extraction effect is obtained.

The invention provides a scenic spot recommendation method based on hybrid supervised learning. The scenic spot recommendation method comprises the steps of obtaining historical tourist touring data;constructing a scenic spot knowledge graph; performing corresponding attribute sub-graph extraction on the scenic spot knowledge graph according to the attribute category of the scenic spot; generating a scenic spot sequence; training the scenic spot sequence and mapping the scenic spot sequence into a low-dimensional vector space to generate a feature vector; adding and averaging the vectors of each scenic spot under different attributes to obtain a fused semantic feature vector of each scenic spot; learning tourist vectors and scenic spot potential vectors; carrying out matrix decompositionon the tourist vector and the fused semantic features to obtain a first interaction vector; obtaining a second interaction vector of the tourist vector and the scenic spot potential vector by using amulti-layer perceptron; splicing the first interaction vector and the second interaction vector and performing normalization processing to obtain a score of the tourist for the scenic spot; ranking the scores of the tourists for the scenic spots from high to low, and obtaining a top _ k scenic spot recommendation list by taking the first K scenic spots with the highest scores.