58 results about "Middle tier" patented technology

This invention discloses a brain-inspired bio-inspired asynchronous event-driven sparse computation method for massive video swarm intelligence, belonging to the field of brain-inspired computing technology. Addressing the problems of resource overload, low computational efficiency, and insufficient system adaptability in city-level video swarm intelligence sensing, it constructs an edge-cloud collaborative asynchronous event-driven architecture: the edge triggers effective events based on a bio-inspired mechanism to generate and transmit sparse video data; the edge uses an SNN-ANN dual-stream architecture to asynchronously extract semantic features; the intermediate layer only triggers spatiotemporal fusion and trajectory stitching for cross-domain events; and the cloud completes global situation aggregation and dynamically adjusts the trigger thresholds of each layer according to system resource load. This invention enables sparse computation with no-event sleep and event-awakening capabilities, significantly reducing data transmission and computational overhead, improving system energy efficiency and stability, and can be widely applied to massive video swarm intelligence sensing scenarios such as smart transportation, security monitoring, and city brains.

This invention belongs to the field of switching chips and relates to a CXL on-chip switching network and CXL switching chip with a multi-target port queue architecture. The switching network consists of several structurally identical CXL basic switching units, employing a three-level cross-connection topology of input layer, intermediate layer, and output layer. Each CXL basic switching unit includes m input ports, n output ports, m CXL ordering modules, and n multi-arbitration modules. The CXL ordering modules implement CXL packet parsing, protocol ordering, and buffered forwarding. The multi-arbitration modules use a round-robin algorithm to implement distributed arbitration. This invention achieves traffic isolation based on target ports, with arbitration complexity increasing linearly and scalable to over 256 ports. Protocol-level ordered switching of CXL packets improves throughput and link utilization while reducing chip area, power consumption, and manufacturing costs. It is suitable for heterogeneous device interconnection in data centers and high-performance computing systems.

This invention provides a flexible production line control method and system based on a large vertical domain model. The method includes: constructing a pruning index combining temporal stability and spatial similarity for the original large vertical domain model to evaluate the redundancy of the activation tensors in the intermediate layers, and performing structured pruning accordingly to obtain a sparse structure model; performing mixed-precision quantization on the pruned model: allocating quantization bits according to the sensitivity of each network layer weight to task loss, and calculating scaling factors and zeros based on the range of non-zero weights to represent sparse weights; calculating the output deviation between the quantized pruned model and the original model under the same input, and training a feedforward network with the former's output as input to fit and compensate for the output deviation; combining the quantized pruned model and the feedforward network to form a composite control model; after deployment, adding the output command of the quantized pruned model to the compensation vector during inference to generate control commands.

The application discloses an operation and maintenance method, device, equipment and program product of a computing power platform, and the scheme comprises the following steps: inputting target operation and maintenance data into a first deep learning model and a second deep learning model to obtain operation and maintenance state prediction information output by the first deep learning model and decision basis information output by the second deep learning model, wherein the decision basis information comprises features and contribution weights of the features on which the first deep learning model relies when generating the operation and maintenance state prediction information; and generating and executing a resource allocation strategy for the computing power platform based on the operation and maintenance state prediction information and the decision basis information. The first deep learning model is trained by using sample operation and maintenance data labeled with real operation and maintenance state labels; and the second deep learning model is trained by using the sample operation and maintenance data and intermediate layer calculation results of the first deep learning model when processing the sample operation and maintenance data as input, and using sample operation and maintenance state prediction information output by the first deep learning model for the sample operation and maintenance data as a supervision signal.

Embodiments of the present application disclose a method and device for stream security prevention and control based on large model generated content. In the embodiments of the present application, the hidden layer state of all intermediate layers when the large model generates the current word element is obtained; the hidden layer state sequence and the current risk state vector of all word elements in any layer of all intermediate layers are determined; the target risk state vector of the current word element is generated according to the hidden layer state sequence and the current risk state vector; and the intervention instruction is generated in response to the target risk state vector being greater than or equal to a set threshold, wherein the intervention instruction is used to instruct the large model to stop generating the current word element or replace the current word element. Through the above method, efficient and accurate risk detection can be performed on the current word element before it is generated.

Computer-implemented method of generating content, method comprising: receiving, from a packages database, one or more sets of package-specific intermediate layers from one or more neural networks configured to generate content, each neural network trained using a respective set of training data, constructing a content generation model by combining package-specific intermediate layers from the one or more neural networks and connecting an output of the combined intermediate layers to one or more pretrained content generation layers; combining of the intermediate layers comprises concatenating outputs of the intermediate layers, constructing the content generation model comprises providing the concatenated outputs to the one or more pretrained content generation layers, the one or more pretrained content generation layers being derived from a pretrained neural network which is configured to generate content from concatenated outputs of intermediate layers, and using the content generation model to generate content for displaying.

This invention discloses a unified data access orchestration and query execution method, a computer device, and a storage medium, belonging to the field of computer technology. The unified data access orchestration and query execution method includes the following steps: S1, determining query parameters; S2, constructing a unified query model; S3, identifying field attribution and modeling query paths; and S4, reasoning and selecting expected paths. This invention achieves a unified structured expression of complex query logic, supports automatic identification of field attribution and path reasoning, and can automatically translate query statements and distribute them to the corresponding databases for execution. It provides an intermediate layer integration mechanism to achieve the final structured return of data. This invention can effectively reduce the complexity of access and management between multiple database systems, improve the flexibility, scalability, and development efficiency of data queries, and is applicable to various scenarios such as data platforms, reporting systems, and data service API platforms.

Provided herein are methods and systems for generating and instantiating an augmented tree model from an artificial intelligence (AI) model via a Parallel and Split Learning (PSL) controller. An example method includes dividing the original AI model into a plurality of partitions. The plurality of partitions include a bottom-level partition including an input layer of the original AI model, a top-level partition including an output layer of the original AI model, and a plurality of middle-level partitions including one or more intermediate layers between the input layer and the output layer. The divided original AI model forms an AI model. The method further includes generating the augmented tree model of the AI model and instantiating one of a plurality of nodes of the augmented tree model into one of a plurality of entities of a network.

This invention relates to the field of data processing technology, specifically to a method for constructing a knowledge base based on multimodal fusion. The method includes: evaluating the information entropy of image and text features at the input layer to determine mutual information between different modalities, and performing dynamic feature pruning. Subsequently, a bidirectional cross-attention mechanism is used in the intermediate layer for deep feature fusion to form collaborative features. These collaborative features are then input to a knowledge extraction unit to extract knowledge triples, and a semantic consistency score is calculated using a cross-modal similarity verification mechanism. Finally, the knowledge triples are filtered based on the score, and knowledge with high reliability is written into a multimodal knowledge graph. This invention, through optimized data preprocessing and structured knowledge filtering logic, ensures the accuracy of the knowledge base content and the stability of the system operation.

A training method includes: an input step of inputting data provenance-specific text data to a large language model in pre-training of the large language model; an acquisition step of acquiring a value of an intermediate layer of the large language model at a time when the large language model is pre-trained with the text data as input; and a training step of generating a training model by supervised learning with the acquired value of the intermediate layer as input data and with provenance information representing provenance of the text data as correct answer data.

This invention provides a method and system for data synchronization across technology stack application systems based on reactive hooks and data chains, relating to the field of distributed system data interaction. The method constructs a data chain hub as an intermediate layer for data synchronization between cross-technology stack application systems, introducing the data chain as an abstract channel for data transmission. In front-end applications supporting reactive hooks, declarative data subscription and automatic updates are achieved by calling reactive hooks. Any connected application system can publish data changes to the data chain, and the hub automatically broadcasts them to all subscribers. This invention employs a version vector mechanism to detect concurrent conflicts and supports multiple conflict resolution strategies, including last-write-wins, automatic merging, and manual resolution. This invention extends the reactive programming paradigm of React to cross-technology stack application systems, solving the problems of high coupling, high development complexity, and lack of conflict detection mechanisms in existing cross-technology stack data transmission.

This invention discloses a practical single-pulse radar training simulator architecture, employing a three-layer structure design consisting of a user layer, a middleware layer, and an execution layer. The user layer, based on a B / S architecture, handles human-computer interaction, training processes, and scenario configuration. It pre-generates and distributes target trajectory-related simulation parameters and pre-set information for abnormal events, while simultaneously collecting operational information and device status and conducting training evaluation. The middleware layer is an embedded control platform used to acquire servo differential angle information in real time, calculate angular error, attenuation, phase, and Doppler echo control parameters, as well as abnormal control quantities, according to the PRF beat. The execution layer is a DRFM simulation sub-unit that receives radar transmitted pulses and applies time delay, Doppler, and amplitude-phase modulation based on control parameters, generating the main target, debris, co-frequency station, and noise echoes for output to the trained radar. This invention achieves a balance between complexity and real-time performance through functional decoupling and separation of responsibilities, supporting high-fidelity training under multiple targets and multiple abnormal conditions.

This application discloses a distributed index construction method, apparatus, and electronic device, relating to the field of distributed storage technology. The method includes: splitting a target key through an intermediate layer proxy to obtain a first key and multiple second keys; wherein the first key is associated with metadata used to determine the second keys, and each second key is associated with a portion of the member data of the target key. Based on the splitting processing logic of the intermediate layer proxy, this application achieves the transformation of the target key into a distributed key by splitting the target key into a first key associated with metadata and multiple second keys each carrying a portion of the member data. This effectively solves the technical problems of thread blocking and uneven cluster memory distribution caused by "large keys" in existing distributed Redis clusters, ensuring the stable and efficient operation of distributed Redis clusters in large data volume scenarios.

The application provides a rural power supply station business carrying capacity evaluation and optimization method. It relates to the technical field of electric power. It is used to solve the problem of distorted evaluation results caused by layer-by-layer transmission of errors. It includes: constructing a hierarchical index library, with the lowest level being specific quantifiable indicators; obtaining the structural weight, subjective weight and objective weight of each quantifiable indicator in the lowest level; determining the comprehensive weight of each indicator from the three weights through game theory, weighting the data of each quantifiable indicator in the lowest level based on the comprehensive weight to obtain the evaluation value of each indicator in the intermediate level; iterating layer by layer until the evaluation value of each indicator in the highest level is obtained; obtaining the business carrying capacity evaluation value based on the evaluation value of each indicator in the highest level; and selecting a power supply optimization strategy that matches the business carrying capacity evaluation value. In each layer of transmission, the method corrects the input bias of the current layer and blocks the transmission of errors to the upper layer through game theory.

The present application relates to a kind of based on virtual-real linkage perception large-scale digital twin model adaptive compression and streaming method, comprising: build edge perception networking, form perception data pool;Construction digital twin basic model;Design layered adaptive hybrid compression algorithm, adopt BP neural network dynamic decision each level model compression strategy, according to core layer, middle layer, edge layer executes different hybrid compression in turn and integrates model, obtains integrated layered compression model;Combined with perception data pool adjustment transmission priority, executes dynamic streaming.This application solves the problem that existing fixed compression algorithm is easy to lose key details or transmission delay is too high by layered adaptive hybrid compression algorithm;Reduce the deployment difficulty and operation cost of large-scale digital twin model;Effectively avoid loading lag, interaction lag and other problems, meet real-time monitoring, remote control and other practical application requirements, further promote the popularization of digital twin technology.

This invention discloses a multi-scale spatiotemporal semantic object state representation method based on brain-inspired biomimetic mapping, belonging to the field of brain-inspired computing technology. Addressing the problems of cross-level semantic abstraction fragmentation, low efficiency of topological association, and semantic redundancy accumulation in city-level video swarm intelligence perception, this invention constructs a hierarchical mapping architecture from edge to middle layer to cloud, sequentially performing progressive abstraction representations from pixel level to individual level, group level, and city level. A multi-dimensional weighted sensor network topology model is constructed, generating a full-cycle continuous spatiotemporal state vector through cross-sensor association matching and trajectory stitching, and completing global optimization and deduplication of cross-camera samples. Simultaneous parallel encoding of metadata and feature streams is performed for swarm intelligence semantic compactness, globally deduplicating and differentially encoding duplicate features and redundant samples. This invention achieves efficient multi-scale spatiotemporal semantic expression through edge-cloud collaboration, significantly reducing transmission and storage overhead and improving the accuracy of cross-regional long-term trajectory stitching.

A four-tuple multi-mode trusted dynamic measurement method for high-reliability fast-response devices, the method establishes a four-tuple measurement model for high-reliability fast-response devices, including subject-object-operation-execution environment, based on the four-tuple measurement model, a hierarchical dynamic measurement architecture is built, the bottom layer is a data support layer, the middle layer is a core measurement layer, and the top layer is a strategy scheduling layer, and the dynamic measurement is initialized after the initialization. The application constructs a classification measurement mechanism from three dimensions of running environment, system behavior and business behavior rules, designs a multi-mode strategy of "trigger event + periodic measurement" by combining business trajectory and system behavior, improves the accuracy and real-time performance of dynamic measurement under the premise of meeting the lightweight demand of embedded devices, ensures the reliable operation of high-reliability fast-response devices, and avoids the security risks of power grid caused by measurement technology defects.

This invention belongs to the field of cloud computing and multi-cloud management technology, and discloses a heterogeneous cloud platform resource management device and method based on a hybrid synchronization mechanism. Addressing the problems of large synchronization latency, high interface pressure, data inconsistency, and tight coupling in existing centralized polling schemes, this invention achieves unified resource management by deploying a proxy middleware layer between the multi-cloud management platform and each cloud platform, and employing a hybrid synchronization mechanism combining event triggering and periodic comparison. The cloud platform proxy module converts raw data into a unified model and maintains a local cache; the event triggering unit monitors internal operations and pushes changes in real time; and the periodic comparison unit periodically identifies differences and synchronizes them through hash comparison. This invention revolutionizes the traditional polling and retrieval mode, achieving near real-time synchronization of resource status; significantly reducing API access pressure and improving data consistency; and, through modular decoupling design, greatly reducing the complexity of system integration and maintenance, and improving scalability and stability.

This application discloses a method, apparatus, device, and storage medium for loading a dynamic library, belonging to the field of computer programs. The method includes: generating assembly language intermediate-level code corresponding to a first function in the dynamic library based on the interface signature of the first function; compiling and linking the source code, linking the calls to the first function in the source code to the assembly language intermediate-level code, obtaining an executable file of the source code; executing the assembly language intermediate-level code when the executable file calls the first function; the assembly language intermediate-level code is used to dynamically load the dynamic library and execute the first function; wherein, the assembly language intermediate-level code is used to cache the parameter context of the function call before dynamically loading the dynamic library and restore the parameter context before executing the first function. This method can improve the generation efficiency of intermediate-level code and reduce the maintenance cost of intermediate-level code.

This application discloses a browser operation method, apparatus, computer device, and readable storage medium, relating to the fields of front-end development and digital medical technology. The code involved in the browser's execution level code package is code that the browser can directly interpret, enabling the browser to utilize its native support for new features without needing Babel as an intermediate layer, resulting in higher execution performance. The method includes: when a request to run a browser is detected, selecting the requesting browser as the target browser and matching the browser level of the target browser; obtaining the level code package associated with the browser level, wherein the code included in the level code package is compiled using the minimum syntax standard supported by the target browser; and controlling the target browser to execute the level code package to ensure successful browser operation.

The application provides a computer virus retrieval method and device, computer equipment and a storage medium, relates to the technical field of computer security, and is used for improving the retrieval efficiency of computer viruses. The method mainly comprises the following steps: establishing a retrieval intermediate layer, and judging whether the virus information of the to-be-detected file is contained in the threat intelligence library based on the retrieval intermediate layer.

The application belongs to the field of natural language processing, and specifically discloses a code generation hallucination mitigation method based on multi-source uncertainty modeling, which comprises the following steps: obtaining token probability distribution of the final output layer of a main model, an intermediate hidden layer and an auxiliary light model in parallel; constructing a multi-source trusted candidate set; calculating direction-aware asymmetric comparison weight, constructing a direction-aware factor according to the probability ratio of the intermediate layer and the final layer; estimating the slope based on the information entropy sequence of the historical decoding steps, and adjusting the timing trend gain; calculating the multi-source comparison decoding score and selecting the token with the highest score. The application also discloses a corresponding system. Through mechanisms such as direction-aware asymmetric comparison, timing trend adjustment and confidence gating, the application can effectively suppress hallucination in code generation without relying on additional training, and improve the grammatical correctness, logical consistency and executability of the generated code.

This technology provides the ability to generate vector representations suitable for searching and classifying documents within a specific domain, without requiring additional training of the language model. [Solution] The search device 1 comprises a transformation unit 12, an adaptation target data storage unit 14, a synthesis unit 15, and a kNN search unit 16. The transformation unit 12 has a BERT model and generates a value vector and a key vector for each domain document, which are stored in the adaptation target data storage unit 14. The value vector is the BERT output, and the key vector is a vector representation generated by the intermediate layer of the BERT model. The synthesis unit 15, in cooperation with the kNN search unit 16, obtains neighboring examples similar to the target document in terms of key vectors from the adaptation target data storage unit 14. Then, by synthesizing the value vectors of the neighboring examples and the value vector of the target document, it generates an adaptation vector, which is the final vector representation of the target document.

This invention relates to the field of generative artificial intelligence and discloses a semantic processing method for 5G message parsing, comprising: inputting a 5G message text sequence into a generative language model; calculating the information entropy of the intermediate network layer; terminating the calculation and extracting the hidden layer state tensor when the information entropy is lower than a threshold; determining the set of legitimate intentions based on the 5G protocol state and transition topology; modulating the initial logarithmic probability tensor using a mask tensor to mask semantic outputs outside the protocol range; calculating the normalized probability distribution sequence and determining the target parsing intention. This invention maps protocol rules to the boundary of the logarithmic probability space, blocks signaling routing failures caused by probability divergence, resolves the conflict between semantic generation and protocol rigor, and ensures the determinism of service parsing under telecommunications conditions.

The application discloses a backdoor defense method of text generation graph diffusion model based on Ollivier-Ricci curvature, and belongs to the technical field of artificial intelligence security. The method constructs a concept database to store text encoder features corresponding to predefined concepts and intermediate layer features of a denoising network; a to-be-detected text prompt is input into a diffusion model to extract text encoder features and intermediate layer features of the denoising network, and reference features are obtained based on semantic similarity matching. A K nearest neighbor graph is constructed based on the text encoder features, Ollivier-Ricci curvature is calculated, and abnormality is determined. If no abnormality is detected, global features are constructed at multiple time steps, curvature is calculated in the same way, and an abnormality proportion is counted. When abnormality is detected at any stage, fine-grained curvature analysis is performed on local intermediate layer features at multiple time steps, and whether a backdoor trigger is contained is determined according to the abnormality proportion, so that backdoor defense is achieved. The method is suitable for backdoor detection of various diffusion models and samplers.

A method (300) includes receiving (305) an input (151) for a machine learning model (110) having a transformer-based architecture with an embedding layer (111), an unembedding layer (113), and three or more transformer layers (112). The method also includes embedding (310) the input at the embedding layer and passing (310) the embedded input from the embedding layer across the transformer layers. Each transformer layer provides one or more feature outputs (114a, 114b). The method further includes obtaining (315) the feature output(s) from a first intermediate transformer layer. The method also includes generating (320) an out-of-distribution (OOD) score (155) for the received input based on the feature output(s) from the first intermediate transformer layer and one or more feature distributions (125a, 125b) from the first intermediate transformer layer. In addition, the method includes comparing (325) the OOD score against a threshold and, in response to determining that the OOD score exceeds the threshold, passing (330) one or more outputs from the unembedding layer to one or more downstream processes (160).

This specification provides a method and system for streaming data processing based on an intermediate layer. The method is applied to an intermediate layer deployed between a server and a third-party large model. The method includes: receiving streaming text data from the third-party large model; identifying resource reference tags based on the streaming text data to determine resource addresses corresponding to the resource reference tags; generating placeholder identifiers corresponding to the resource addresses and replacing the resource reference tags with placeholder tags, the placeholder tags including the placeholder identifiers; forwarding the placeholder tags and the streaming text data to a client; wherein the client is configured to display the streaming text data in real time on a display page, and to locate and update placeholder elements corresponding to image positions on the display page based on the placeholder tags.

In one embodiment, software is operable when executed to receive a request for a video fragment from a CDN by a mid-tier video transformer operating between the CDN and servers, determine a type of the request, and identify configuration rules among configuration rules are centralized in the mid-tier video transformer based on the type of the request to transform the request. The software is further operable when executed to determine a media type of the video fragment based on the configuration rules, identify a server based on the media type, and route a transformed request to the server by the mid-tier video transformer.