88 results about "Large networks" patented technology

A network operation and maintenance metrics question and answer method and system based on a knowledge graph and a large model, relating to the technical field of computers. According to the method, a large network model may be triggered to extract at least one type of target operation and maintenance data from entity objects, entity relationships, entity attributes, and operation and maintenance events in network operation and maintenance data, so as to construct a knowledge graph on the basis of the target operation and maintenance data; the knowledge graph can assist the large network model in responding to knowledge question and answer, and the knowledge question and answer result can further assist in enhancing a modeling process of the knowledge graph, thereby improving the response precision of the large network model to knowledge question and answer in the field of operation and maintenance; and the knowledge graph can model and represent operation and maintenance actions, thereby achieving the standardization and automation of the operation and maintenance actions. The large network model can further optimize these operation procedures by learning a large amount of operation and maintenance action data, thereby improving the efficiency of operation and maintenance.

Systems, methods, and computer readable media related to: training an encoder model that can be utilized to determine semantic similarity of a natural language textual string to each of one or more additional natural language textual strings (directly and / or indirectly); and / or using a trained encoder model to determine one or more responsive actions to perform in response to a natural language query. The encoder model is a machine learning model, such as a neural network model. In some implementations of training the encoder model, the encoder model is trained as part of a larger network architecture trained based on one or more tasks that are distinct from a “semantic textual similarity” task for which the encoder model can be used.

The invention discloses a forest fire smoke monitoring method and related device based on false detection feedback and model self-evolution, and the method comprises the steps: obtaining a video stream of a monitoring region, carrying out the preprocessing, inputting a target detection model and a scene understanding model, and judging whether a candidate alarm event is generated or not through consistency analysis; when a candidate alarm event is generated, a work order is generated for manual judgment, a real alarm executes a disposal process, and a false alarm is stored in a database; screening high-value false detection samples based on an information entropy priority strategy, storing the samples into a dynamic memory bank, selecting the samples to form an incremental training set, performing incremental training on a pre-training teacher model, and introducing low-rank constraint stability parameters; migrating the discrimination ability of the teacher model to the student model through knowledge distillation, wherein alignment of a channel dimension and a multi-scale space dimension is carried out; and redeploying the updated student model to execute target detection. The invention aims to solve the problems that the current monitoring system depends on manpower, the coverage capability is limited, the response is slow and the cost is high, and the traditional method is high in false alarm rate, easy to leak and misjudge, difficult in large-scale network edge deployment, limited in light-weight model discrimination capability, lack of linkage between alarm processing and model updating and the like, and more accurate and efficient forest fire smoke monitoring is realized.

Efficient scaling of in-network compute operations to large numbers of compute nodes is disclosed. Each compute node is connected to a same plurality of network compute nodes, such as compute-enabled network switches. Compute processes at the compute nodes generate local gradients or other vectors by, for instance, performing a forward pass on a neural network. Each vector comprises values for a same set of vector elements. Each network compute node is assigned to, based on the local vectors, reduce vector data for a different a subset of the vector elements. Each network compute node returns a result chunk for the elements it processed back to each of the compute nodes, whereby each compute node receives the full result vector. This configuration may, in some embodiments, reduce buffering, processing, and / or other resource requirements for the network compute node or network at large.

The application relates to the technical field of network communication, and discloses a network self-healing device based on node state data, which aims to improve the stability and self-healing ability of network communication. By monitoring the network state in real time, combining the extended Kalman filtering algorithm to perform state estimation and filtering, and combining the ant colony algorithm and the simulated annealing algorithm to perform fault prediction, path selection and resource allocation, the potential fault can be predicted and quickly recovered. The device comprises five modules of state monitoring, state estimation and filtering, fault prediction and prevention, fault recovery and resource reallocation, forms a closed-loop optimization mechanism, and is especially suitable for complex large network environments, such as data center networks.

The present application relates to a method for individual identification of electromagnetic signals using open set semi-supervised learning technology, belonging to the field of electromagnetic signal identification processing. The technical scheme of the present application trains a large classification network as a first model through self-supervised pre-training, so that it can more fully learn the internal relationship of unlabeled data and improve the network feature extraction capability; a small-scale classification network model closer to the characteristics of the identification sample is used as a second model, and in semi-supervised learning, the feature extraction capability of the pre-trained model is used to filter out unlabeled data other than the target classification, reducing the negative impact of out-of-distribution unlabeled data, and at the same time, model distillation is used to enable the small-scale classification network model to learn the generalization and robustness of the large network model for data features. Through the above three aspects, the influence of data other than the target classification on the semi-supervised learning network recognition accuracy in real situations is reduced, and robust identification of electromagnetic signals is achieved.

The application belongs to the field of FPGA-based hardware accelerator design, and proposes a point cloud target detection neural network accelerator and an acceleration method based on FPGA. The convolution accelerator comprises an instruction decoding module, a convolution module, an accumulation module, a truncation module and an activation function. The intermediate value cache only temporarily stores the channels of one feature map, consumes less BRAM, and is suitable for accelerating the point cloud target detection neural network with large feature map size, large number of channels and large network size. The application has flexible configurability and excellent reusability, so the application can accelerate different convolution layers of the point cloud target detection neural network, and is suitable for being integrated as an IP into a large design, reducing FPGA resource consumption and improving resource utilization. Moreover, the weight and the input feature map share one data input channel, and the convolution intermediate value is temporarily stored on-chip BRAM, reducing the feature map carrying frequency and saving the FPGA interface bandwidth.

The present application is a kind of lightweight target detection method for embedded platform, which overcomes the problems of large network parameter quantity, slow detection speed and poor precision index in the prior art. The present application not only solves the problems of slow speed and low accuracy of the current detection method in the deployment of embedded devices, but also optimizes the computation graph for specific operators from the hardware level, enabling fast and accurate target detection on limited resource devices. The present application comprises the following steps: step 1: obtaining basic data and making data set; step 2: data enhancement preprocessing; step 3: constructing a benchmark network model and pre-training; step 4: building a lightweight detection network model; step 5: reparameterization operation and pruning on the overall network structure; step 6: knowledge distillation to restore accuracy; step 7: deployment and acceleration of embedded platform.

The application provides a photonic neural network system based on a multi-task neural network, which comprises: an optical computing module, taking a photonic artificial intelligence chip as a core, modulating an optical signal, and completing multiplication, summation and nonlinear operation of high-precision and high-speed optical analog quantities; a multi-task neural network module, having a main branch neural network and multiple secondary branch neural networks, and capable of simultaneously processing multiple classification tasks and regression tasks; and an optical computing communication and control module based on FPGA, connected with the optical computing module and the multi-task neural network module, and used for realizing real-time and high-speed communication of the optical computing module and the multi-task neural network module. The system has the characteristics of high bandwidth and low energy consumption, utilizes high-dimension parallel computing to greatly reduce the time of original neural network serial operation, has the advantages of large network scale and fast parallel operation speed, and can improve recognition accuracy and shorten training time on the basis of a single task.

The application relates to the field of electric communication technology, in particular to a multi-thread processing method based on a core surface target IP address, which comprises the following steps: establishing a signaling surface data packet according to a service request of a user, wherein the signaling surface data packet comprises a base station IP of a user terminal, signaling data and a target IP, the signaling data comprises a TEID of the user terminal; sending the signaling surface data packet to the base station; sending the signaling surface data packet to an edge distribution gateway by the base station; sending the signaling surface data packet to a signaling collection node for identification by the edge distribution gateway, and selecting a corresponding thread according to an identification result and transmitting the thread to a core network for data transparent transmission of a large network user; the application adopts a scheme of resource allocation by using a core network IP in a signaling learning library, ensures that a same user terminal is only processed in a same transmission thread, and avoids the problem of efficiency reduction caused by switching and data calling of different transmission threads.

In a transmission scene with large network delay or large out-of-order distance, in order to solve the problems that the bit width of a packet receiving bitmap of an RDMA data receiving end is too large and storage resources are occupied too much, the invention provides a packet receiving bitmap management method of out-of-order RDMA. According to the method, an RDMA queue does not apply for a storage space in advance for an annular packet receiving bitmap of Q), the annular packet receiving bitmap is segmented into a plurality of sub-bitmaps, and a sub-bitmap space is dynamically applied or released from an available sub-bitmap address pool according to needs; the available sub-bitmap address pool is a queue, all sub-bitmap addresses are queued into the address pool during initialization, when the QP needs a sub-bitmap, one sub-bitmap address is dequeued from the address pool, and after a certain sub-bitmap of the QP is used, the used sub-bitmap address is queued back into the address pool; when a certain sub-bitmap of the QP is 1, the sub-bitmap space can be released in advance. According to the invention, the RDMA data receiving end supports the packet receiving bitmap with a large bit width, the occupied bitmap storage resources are small, and the utilization rate is high.

The present application relates to the technical field of pavement crack detection, and particularly relates to a lightweight pavement crack detection method, wherein a lightweight feature extraction module SandGlass and a coordinated attention mechanism are integrated into an encoding-decoding structure to construct a lightweight pixel-level crack detection network, a pavement image database is used to train the network model, so that the network model can realize the same detection effect as a larger network model under less computing power, the coordinated attention mechanism module suitable for mobile terminals is introduced to reduce the influence of noise and improve the accuracy of crack detection, and finally the encoding-decoding structure is improved to fuse feature maps of different levels, so that the final prediction result is closer to the actual pavement condition. Further, the improved method can be effectively applied to mobile devices to realize real-time and effective end-to-end detection.

The present application relates to the technical field of image processing, and more particularly to a water surface garbage identification method based on deep learning, comprising cutting pictures of a water surface garbage identification dataset; after adding a residual connection between a feature map and an input graph, an inhibition weight is obtained; each channel of the input graph is multiplied by the inhibition weight after subtraction to obtain a weighted feature map; two deep separable convolution groups are used in the deep layer to form a DSC module, and a Res module is used in the shallow layer; a deep separable convolution combined with a channel attention mechanism is used to obtain a one-dimensional feature vector, and a one-dimensional convolution is used to generate a channel weight, and finally the channel weight is multiplied by the original feature map; an ECA module is added after the Res module and the DSC module of the encoder; and a deep separable convolution module is used in the decoder. The present application solves the problems of noise interference such as reflection and reflection in the water surface scene, large network structure model size, large storage space occupation and high consumption of computing resources.

The application discloses a similar network behavior user identification method based on an unsupervised variational autoencoder, and relates to the technical field of network security. The method is based on super-large network communication metadata, uses an unsupervised variational autoencoder model, learns deep law features of user network behaviors from massive user network flow metadata, abstracts network behaviors of users into high-dimensional vector representations of essential behavior modes, and finally identifies users by using the similarity between vectors. Specifically, the method comprises network flow metadata, network behavior statistical feature extraction, behavior feature image processing, deep learning model training, deep feature extraction and similar behavior user identification.

This application belongs to the field of interconnection network reliability and fault diagnosis technology, and discloses a method for intermittent fault diagnosis of low-bandwidth long interconnection networks based on graph attention mechanism. Targeting the hierarchical recursive structure and high connectivity of the network, under the PMC fault diagnosis model, a multi-round testing strategy is used to obtain test symptoms within the node's neighborhood. A feature vector is constructed for each node using local statistical feature extraction methods, and preprocessed using zero-padding and noise enhancement techniques. Finally, a graph attention network model is constructed, and the importance weights of neighboring node test results are dynamically learned using the attention mechanism to achieve accurate diagnosis of node fault states in the network. This application leverages the powerful local information aggregation capability of graph attention networks to overcome the diagnostic limitations of traditional algorithms, maintaining high diagnostic accuracy and robustness even with high fault rates, incomplete test symptoms, and large network sizes.

The present application provides a network topology configuration method, a communication system, and a related device. In the method, before distributed training of an AI model begins, a network controller can analyze a training task and determine traffic demand information for the training task. The traffic demand information comprises a bandwidth demand between different processors in a training process of an AI training task. Network topology information of an optical-electrical hybrid network can be determined on the basis of the traffic demand information. Configuration information of an optical switching device is then generated on the basis of the network topology information, so that the optical switching device can perform link switching on the basis of the configuration information and complete topology reconfiguration of the optical-electrical hybrid network. In this way, the network controller can actively perform traffic planning before AI model training, avoiding the problem of large network delay in the training process caused by a slow link switching speed of the optical switching device.

The embodiment of the invention provides a session establishment method, device and system, and the method comprises the steps that a first network element receives first information from a second network element, and the first information is used for requesting to establish a first session; sending second information to a third network element, wherein the second information is used for requesting to obtain subscription data and / or a strategy of the terminal; receiving third information, wherein the third information comprises subscription data and / or a strategy of the terminal; and selecting a fourth network element according to the third information, and sending fourth information to the fourth network element, the fourth information being used for requesting to establish a first session, the first network element being a network element providing a session management function, and the first network element being used for managing or controlling one or more sessions of the terminal. By adopting the means, the problem that the AMF is frequently upgraded and the base station is influenced due to the fact that new services relate to SMF selection strategy modification is solved, and the SMF upgrading problem that large network SMF transformation needs to support selection of a private network SMF in a dual-domain private network scene is solved.

The invention relates to a media distribution method and device, computer equipment, a readable storage medium and a program product. The method comprises the steps of detecting a network type of a terminal in response to a media acquisition request sent by the terminal for a target media; according to whether the network type of the terminal is a preset network type, sending the first index file or the second index file to the terminal; wherein the first index file is used for obtaining fragment links corresponding to all first media fragments of the target media, and the second index file is used for obtaining fragment links corresponding to all second media fragments of the target media; the duration of the first media fragment is less than that of the second media fragment. By adopting the method, the timeliness of code rate switching in the streaming media transmission process can be improved, and the phenomenon of media playing lagging under the condition of large network fluctuation is reduced.

The application provides an information transmission method and device, equipment and a non-transient readable storage medium. The information transmission method comprises the following steps: first information is directly sent to a network function device by calling a service of the network function device, the first information being application level information between a radio access network device and the network function device; and / or second information sent by the network function device by calling a service of the radio access network device is received, the second information being application level information between the network function device and the radio access network device. The present scheme can support direct information interaction between the radio access network device and the network function device through a service interface, avoid forwarding through an intermediate node, shorten a transmission path, reduce signaling routing delay and network overhead, and well solve the problems of large signaling routing delay and large network overhead in the prior art.

Provided in the embodiments of the present application are a method for determining a transport network layer address of a base station, and an apparatus, which are applied to base stations. The method comprises: acquiring transport network layer addresses associated with a target ground area; and when the base station operates in the target ground area, determining a target transport network layer address associated with the target ground area as the transport network layer address of the base station. The present invention solves the problem of large network overheads caused by excessive signaling interaction in non-terrestrial networks.

The application discloses an edge-computing-based security camera interference target filtering system, and belongs to the technical field of intelligent security protection. The system comprises an edge computing module, a multi-modal feature extraction module, an adaptive feature fusion module, an interference target classification module, a real-time filtering execution module and an edge-cloud collaborative optimization module. The edge computing module is arranged at the security camera end and integrates a lightweight neural network processing unit. The multi-modal feature extraction module simultaneously extracts visual features, motion features and environment perception features. The adaptive feature fusion module dynamically allocates the weights of different modal features by adopting an attention mechanism. The interference target classification module constructs a multi-level classification system. The real-time filtering execution module adopts a three-level filtering strategy. The edge-cloud collaborative optimization module realizes global optimization and parameter updating of the model. The application effectively solves the problems of low single-feature recognition accuracy, large network delay and fixed threshold value that cannot adapt to scene changes in the prior art.

The application discloses a resource pushing method and device, equipment, a storage medium and a computer program product. The method comprises the following steps: calling a first interaction prediction network to predict an interaction probability of a to-be-pushed object performing a target interaction on a to-be-pushed resource; determining a matching degree between the to-be-pushed object and the to-be-pushed resource based on the interaction probability; if the matching degree meets a pushing rule, pushing the to-be-pushed resource to the to-be-pushed object; the first interaction prediction network is trained based on a second interaction prediction network with a larger network scale, a first sample resource and a historical short-time interaction sequence related to the first sample resource; the second interaction prediction network is trained based on the first sample resource and a historical long-time interaction sequence related to the first sample resource, and the historical long-time interaction sequence comprises the historical short-time interaction sequence; the first interaction prediction network and the second interaction prediction network are both used for target interaction prediction; and the pushing accuracy and the pushing timeliness during resource pushing can be considered simultaneously.

The embodiment of the invention provides a wide-area computing power cascade network architecture based on hierarchical Dragonfly + Fat-Tree, and the network architecture comprises at least one computing power cluster which comprises a plurality of computing power chips in communication connection; the at least one first network architecture comprises a plurality of first nodes, and the first nodes in the same first network architecture are in communication connection to form a tree topology structure; the first-level autonomous architecture comprises a plurality of second nodes, the second nodes in the same first-level autonomous architecture are in communication connection to form a direct connection topological structure, and different first-level autonomous architectures belong to different areas; wherein at least one computing power cluster is in communication connection with a first-stage autonomous architecture through a first network architecture, and the first-stage autonomous architecture is connected with the computing power cluster in an area to which the first-stage autonomous architecture belongs. Therefore, according to the embodiment of the invention, a super-large-scale network architecture can be constructed.

The invention discloses a crane remote control image delay detection method based on machine learning, which belongs to the technical field of crane remote control, and comprises the following steps: S1, data acquisition and preprocessing; s2, feature extraction; s3, training a machine learning model; s4, carrying out real-time image delay detection; and S5, adjusting the control strategy. According to the method, a machine learning algorithm is utilized, complex features and time sequence relations in image data can be automatically learned, compared with a traditional method, image time delay can be more accurately detected, the detection precision is effectively improved, reliable data support is provided for remote control of the crane, and by continuously updating training data and optimizing a model, the accuracy of remote control of the crane is improved. The method can adapt to different network environments, crane operation scenes and working condition changes. Good detection performance can be kept no matter in an environment with large network fluctuation or in a complex operation scene, and the universality and practicability of the method are improved.

This invention discloses a 5G MIMO combined antenna and terminal. The 5G MIMO combined antenna includes a substrate, a first antenna element, a second antenna element, and an isolation slot. A conductive layer is disposed on one side of the substrate. The first and second antenna elements have identical structures and each includes a first radiating part, a second radiating part, a third radiating part, and a feed part. The isolation slot is disposed near the centerline of the conductive layer in the width direction. The first and second antenna elements are symmetrically arranged about the isolation slot. This 5G MIMO combined antenna supports all current 5G frequency bands (sub-6GHz bands) globally, except for millimeter wave bands. It has the advantages of flexible operating frequency bands and large network capacity. The antenna structure is simple and compact, with good isolation between array elements, and good antenna electrical performance within the operating frequency band.

To obtain a larger network capacity by setting three network hierarchies in a Clos network.SOLUTION: The network designed by the network design device 1 is provided with a plurality of first stage switches belonging to a first stage and connected to a plurality of terminals, a plurality of second stage switches belonging to a second stage and connected to the plurality of first stage switches, and a plurality of third stage switches belonging to a third stage and connected to the plurality of second stage switches. The network design device 1 includes a calculation unit 12 that calculates the number of terminals and the number of first stage switches so that the total probability of each blockage occurrence probability in a case where the number of terminals is increased by one from the maximum value of the number of terminals when the network is non-blocked does not exceed the allowable blockage rate ε and the network capacity is maximized in a network that satisfies the number of ports N of each switch, the total number a of a plurality of switches, and the terminal utilization rate p.SELECTED DRAWING: Figure 8

Devices, systems, methods, and processes for dynamically controlling power supplied from power supply units (PSUs) and battery units of network devices are described herein. Generally, network devices rely on redundant power supplies or large hold-up capacitors to address power supply issues. Redundant power supplies lead to sub-optimal efficiency, while hold-up capacitors lead to bulky network device taking up space within the PSUs. Therefore, the present disclosure describes disposing one or more battery units in linecard slots or PSU slots. The battery unit may provide redundancy for the power supply and can act as an effective filter for power signal fluctuations. The battery units and the PSUs are dynamically controlled based on a load demand associated with the network device and power supply sources connected to the network device. Thus, the battery unit actively participates in load sharing with the PSUs to operate the PSUs more efficiently.

The application provides a network path optimization method, device, equipment, storage medium and program product, wherein the method comprises the following steps: constructing a network topology based on a communication network, obtaining a routing request, determining a source port and a target port, and pruning the network topology; obtaining sensing data of each link, wherein the sensing data comprises a time delay, performing feature extraction on the time delay of each link, inputting the feature vector of the time delay of each link into a pre-constructed prediction model to obtain a predicted time delay of each link; generating a network topology graph based on the pruned network topology, performing global path search in the network topology graph based on the sensing data of each link and the predicted time delay of each link, and determining an optimal path. The application comprehensively considers multiple indexes for path optimization, has high accuracy and strong reliability, can maximize the improvement of network communication quality, and is suitable for complex large networks.

The application discloses an edge computing method based on artificial intelligence, and particularly relates to the technical field of edge computing and artificial intelligence. The method comprises the following steps: deploying a light neural network model on an edge device to realize local data processing; dynamically optimizing the allocation strategy of a computing task between an edge node and a cloud end through a deep reinforcement learning algorithm; and realizing cross-device collaborative training by using an improved federated learning framework to guarantee data privacy. Through real-time data processing on the edge side, the average system response delay is reduced by more than 40%; the task offloading algorithm based on Q-learning can reduce network traffic by more than 30%; and the innovative differential privacy federated learning mechanism reduces the data leakage risk by 60% on the premise of ensuring model accuracy. Through multi-level technical cooperation, the application effectively solves the problems of high response delay, large network bandwidth occupation and insufficient data privacy protection in the traditional cloud computing architecture.