221 results about "Network embedding" patented technology

For a multi-tenant environment, some embodiments of the invention provide a novel method for (1) embedding a specific path for a tenant's data message flow through a network in tunnel headers encapsulating the data message flow, and then (2) using the embedded path information to direct the data message flow through the network. In some embodiments, the method selects the specific path from two or more viable such paths through the network for the data message flow.

For a multi-tenant environment, some embodiments of the invention provide a novel method for (1) embedding a specific path for a tenant's data message flow through a network in tunnel headers encapsulating the data message flow, and then (2) using the embedded path information to direct the data message flow through the network. In some embodiments, the method selects the specific path from two or more viable such paths through the network for the data message flow.

A three-dimensional (3D) chip stack structure and method of fabricating the structure thereof are provided. The 3D chip stack structure includes a plurality of vertically stacked chips which are interconnected and bonded together, wherein each of the vertically stacked chips include one or more IC device strata. The 3D chip stack structure further includes an air channel interconnect network embedded within the chip stack structure, and wherein the air channel interconnect network is formed in between at least two wafers bonded to each other of the vertically stacked wafers and in between at least two bonded wafers of the vertically stacked wafers at a bonding interface thereof. In addition, the 3D chip stack structure further includes one or more openings in a peripheral region of the chip stack structure that lead into and out of the air channel interconnect network, so that air can flow into and out of the air channel interconnect network through the one or more openings to remove heat from the chip stack structure.

Methods and systems for embedding a network in a latent space include generating a representation of an input network graph in the latent space using an autoencoder model and generating a representation of a set of noise samples in the latent space using a generator model. A discriminator model discriminates between the representation of the input network graph and the representation of the set of noise samples. The autoencoder model, the generator model, and the discriminator model are jointly trained by minimizing a joint loss function that includes parameters for each model. A final representation of the input network graph is generated using the trained autoencoder model.

The invention discloses a social information recommendation algorithm and system based on a deep network embedding feature. A network embedding model is trained on the social network of a user in advance to extract network feature representation of the user, the extracted network feature is integrated into a matrix decomposition model; and a final generated model is used for score prediction and project recommendation. The social information recommendation algorithm and system based on the deep network embedding feature in the invention can not only make in-depth use of the social network information, but also can use a collaborative filtering model for recommendation as well as have better information recommendation accuracy and convergence.

The invention presents a method for identifying authors based on heterogeneous embedding network model. The method for identifying authors mainly comprises the following parts, to be specific, node embedding, heterogeneous embedding network model, shared embedding, joint training, and inputting paper to recognize and identify authors. The process comprises the following steps, firstly, inputting anonymous papers, analyzing the papers to identify key information and constructing characteristic representations; utilizing a heterogeneous embedded network model of task-guide teaching and path strength, including a specific embedded sub model of a specified task and a general embedded sub model of the path strength to generate a joint target and then performing a joint training; and finally confirming a rank for possible authors and outputting the highest ranking, namely a real author. The method for identifying authors based on heterogeneous embedded network model breaks through the limit that the heterogeneity of network is ignored and only isomorphic networks can be handled in the conventional embedding due to the fact of utilizing the universal network embedding which is independent of a specific task. By utilizing the network embedding with the guide task and strengthened path, the method is more efficient when applied to recognizing the real author, as compared with the existing method.

The invention provides a socialized recommendation method based on a matrix decomposition and network embedding joint model. The method comprises the following steps: constructing a user-article scoring matrix and user-user social network and generating a user social corpus according to user-user social network; utilizing a user-article scoring data and a user social corpus training matrix decomposition and network embedding joint model to obtain a user feature matrix and an article feature matrix; predicting an unobserved score according to the user feature matrix and the article feature matrix; and recommending a plurality of articles with relatively high prediction score values to the corresponding users. According to the method, a matrix decomposition model and a network embedding model are seamlessly integrated by designing a unified target function; based on a unified optimization framework, bidirectional promotion and collaborative optimization between a matrix decomposition model and a network embedding model are realized, so that interested articles can be accurately recommended to a user.

The invention belongs to the technical field of social network user identification, and discloses a cross-social network user identity identification method based on a community structure, which comprises the following steps: step a, respectively mapping a source network and a target network to a low-dimensional vector space by adopting a network embedding mode; and step b, based on the vector space, training a BP neural network in a supervised manner to obtain a BP neural network model for realizing mapping from a source network to a target network, and identifying the user identity of the source network in the target network through the BP neural network model. In the process of learning the feature vector representation of the social network nodes, the proximity features and the community structure features of the nodes are fused, the structure features of the social network are reserved to the maximum extent, and the user identity recognition accuracy is improved.

The invention relates to the technical field of recommendation, in particular to a meta-path-based network embedded movie recommendation method, which comprises the following steps of: obtaining information of a user and a movie; constructing the information of the user and the movie into a heterogeneous information network; obtaining a node sequence of each node by using random walk based on a meta-path; learning the sequence of each node through a skip-gram model to obtain network embedding vectors under different meta-paths; fusing the embedded vectors of the user and the movie to obtain afused embedded vector, fusing the fused embedded vector into the scoring preference, and calculating scores of the user for the movie possibly interested in; and recommending the similar users and movies to the users according to the scores. According to the invention, users and movies can be connected through a heterogeneous information network, and more information between more users and moviescan be obtained through meta-path-based network embedded representation, so that the recommendation accuracy is improved.

Methods and systems for determining a network embedding include training a network embedding model using training data that includes topology information for networks and attribute information relating to vertices of the networks. An embedded representation is generated using the trained network embedding model to represent an input network, with associated attribute information, in a network topology space. A machine learning task is performed using the embedded representation as input to a machine learning model.

The invention belongs to the technical field of computer vision, deep learning and medicine, particularly relates to a non-contact heart rate measurement method, system and device based on a face image, and aims to solve the problems that an existing non-contact heart rate measurement method based on an image is greatly influenced by ROI selection, environment and other factors, and the measurement accuracy is low. The heart rate calculation error rate is large, and the real-time performance is low are solved. The method comprises the steps of obtaining a face position from a face video through face key point detection and positioning, and extracting a face local ROI area frame by frame as network model input; on the basis of a convolution and time sequence network cascade model, dividingheart rate intervals into different interval categories, embedding a channel attention network SENet into a convolution module, weights are learned according to the channel importance degree, and finally acquiring the heart rate interval categories corresponding to input videos. CNN feature extraction and the LSTM long-short-term memory neural network are combined, and the channel attention network is embedded, so that heart rate non-contact measurement with low error rate and high efficiency is realized.

Embodiments of the present specification provide a method of training an embedded model of a relational network, comprising obtaining a calibration node from the relational network, each calibration node having a corresponding tag value to show a credit rating of a corresponding user; then, the node characteristics of the calibration node, the neighbor node set, and the edge characteristics of each connection edge with each neighbor node being determined; determining a node embedding vector of the primary iteration of each calibration node and an edge embedding vector of the primary iterationof each connection edge based on the node feature, the edge feature and the first parameter set; a multi-level vector iteration beign then performed to determine a node embedding vector for the multi-level iteration of each calibration node; further, a prediction value of the node being determined based on a node embedding vector of a multi-level iteration of the calibration node and a set of prediction parameters; finally, each parameter value being adjusted to minimize the loss function, which is determined based on the prediction value and label value of each calibration node.

The invention realizes a set of differential privacy recommendation method based on heterogeneous information network embedding. The differential privacy recommendation method comprises the followingfour steps of: performing network representation learning by using HAN, and calculating heterogeneous attention sensitivity by using characterizations of HAN and an attention weight result; based on adifferential privacy definition, using the heterogeneous attention sensitivity to generate corresponding random noise, and generating a random noise matrix through using a heterogeneous attention random disturbance mechanism; constructing an objective function of differential privacy recommendation embedded with heterogeneous information for learning to obtain a prediction score matrix; and outputting the score matrix as a prediction score capable of keeping privacy. Therefore, the original scoring data is protected for the recommendation system scene under the heterogeneous information network, an attacker is prevented from improving the reasoning attack capability by utilizing the heterogeneous information network data acquired by other channels, and the original scoring data can be guessed or learned again with high probability by observing the recommendation result change of the score.

The invention discloses a semi-supervised network representation learning algorithm LSDNE (Labeled Structural Deep Network Embedding) based on a deep compression self-encoder. The method comprises thefollowing steps: building a model, pre-training the input data with a deep belief network (DBN) to obtain the initial values of the model parameters, and taking the adjacency matrix and Laplace matrix of the network as inputs; encoding the network by a self-encoder with deep compression, and obtaining the global structure of the node; using Laplacian feature mapping, and obtaining the local structural features of nodes; using an SVM classifier to classify the known label nodes and optimize the whole model; using the Adam optimization model and obtaining a representation of the node. The invention can simultaneously use the structure information of the network and the label information of the node to carry out network representation learning, and the deep learning model is used, so that the performance of the representation of the node on the label classification task is better than the existing algorithm. Deep compression self-encoder can reduce the over-fitting phenomenon and make the model have better generalization performance.

A sensor network embedded operation system based on the priority scheduling belongs to the technical field of wireless sensor network, and comprises a task priority setting interface module, a task scheduling configuration assembly module, and a priority-based task scheduling module, wherein the task priority setting interface module is connected with an application program and transmits the taskpriority information; the task scheduling configuration assembly module is connected with a system kernel and transmits the task scheduling information; and the priority-based task scheduling module is connected with the task scheduling configuration module and transmits the task priority scheduling information. By adopting a priority-based multi-task preemptible scheduling strategy for kernel scheduling, the system can provide a priority scheduling interface for the application program, and can assign a priority to the task before the task is added to a queue. Therefore, the system can schedule the tasks according to the assigned priority, and the task with higher priority is executed firstly and then is added to the queue.

Detecting information relating to a medium in an interactive media device. The device includes a surface able to receive the medium and a sensor network embedded within the surface. The sensor network includes a configuration able to detect medium information. The device has a processor able to issue signals to the sensor network and to receive signals indicating capacitance change caused by proximity of a region of the medium to the sensor network. Thus, the processor is able to detect medium information based on the capacitance change.

The invention discloses a network link prediction method based on multiple semantic influences of multiple neighbor nodes, relates to data mining and topological structure analysis, and belongs to a research problem in the field of social computing. The method comprises the following steps: data analysis: analyzing node interest characteristics and network structure characteristics based on node behaviors and node relationship data in a social network; and model training: obtaining an embedded vector of each node by the model in combination with multiple semantic influences of multiple neighbor nodes; predictive analysis: measuring the probability of existence of friend links by using the similarity between the embedded vectors of the node pairs. According to the method, the constant influence scores of the neighbors are not used, and the special semantic influence of each neighbor on the node is simulated. Local-level and global-level semantic influences of neighbor nodes in network embedding training are jointly simulated, and a joint embedding vector based on the semantic influences of all the neighbor nodes is trained for each node.

Embodiments of the invention provide a method, system and computer program product for embedding a global barrier and global interrupt network in a parallel computer system organized as a torus network. The computer system includes a multitude of nodes. In one embodiment, the method comprises taking inputs from a set of receivers of the nodes, dividing the inputs from the receivers into a plurality of classes, combining the inputs of each of the classes to obtain a result, and sending said result to a set of senders of the nodes. Embodiments of the invention provide a method, system and computer program product for embedding a collective network in a parallel computer system organized as a torus network. In one embodiment, the method comprises adding to a torus network a central collective logic to route messages among at least a group of nodes in a tree structure.

The invention discloses a multi-mode depth network embedding method for fusing structure and attribute information, and relates to the technical field of complex network analysis. The method comprisesthe steps of establishing a network adjacency matrix and an attribute matrix, performing preprocessing, serially inputting structural features and attribute features into an encoder and a decoder, outputting a reconstructed adjacency matrix and an attribute matrix, updating a parameter iterative calculation and the like, and finally taking the output of the encoder as a final node representation.Based on the depth learning method, the invention can overcome the shortcoming that the existing shallow linear method is difficult to depict the highly nonlinear structure of the network, can map nodes in the network to a low-dimensional embedded space, and effectively maintains the structural characteristics and attribute characteristics of the nodes.

The invention discloses an article score prediction method based on matrix decomposition and neural collaborative filtering. The method comprises the following steps: (1) constructing a user-; An article scoring matrix; (2) performing matrix decomposition on the training matrix; (3) constructing a neural collaborative filtering network; (4) performing neural collaborative filtering on the trainingmatrix; (5) extracting user and article characteristics of the neural collaborative filtering network embedded layer; (6) constructing a nearest neighbor feature matrix; (7) generating a scoring training set and a scoring test set; (8) training the fully connected neural network; and (9) carrying out score prediction on the score test set. The method has the advantages that the user and article information is fully mined, the article score prediction accuracy is high, and the expandability is high.