89 results about "Reasoning system" patented technology

The invention belongs to the related technical field of product detection, provides a multi-view zero sample anomaly detection method and system based on cross-modal prompt reasoning, and aims at solving the problem of multi-view zero sample anomaly detection by constructing core technologies such as multi-view pose estimation and alignment, static-dynamic prompt collaboration, vision-language progressive fusion, feature space semantic enhancement and the like. And a set of end-to-end anomaly detection and reasoning system is formed. Particularly, a collaborative mechanism of a dynamic learnable prompt pool and a static attribute prompt library is designed, deep fusion of prompts is realized through cross attention, and multi-view feature compression and semantic decoding are performed by adopting a visual angle self-adaptive hybrid expert model. Zero sample anomaly detection and visual question and answer performance is further improved on multiple industrial public data sets, and the method can be widely applied to industrial precision part quality inspection, intelligent manufacturing and other complex scenes needing high-precision and multi-view perception and semantic reasoning.

The invention relates to the technical field of artificial intelligence and data processing, and discloses a multi-mode identity relation inference system based on a graph neural network. The system comprises a multi-modal feature extraction module, a cross-modal alignment module, a graph structure construction module, a dynamic relation reasoning module and a decision output module. According to the method, the cross-modal alignment module is introduced to project the image features and the text features to a unified public semantic space, so that the nonlinear distribution difference of heterogeneous modals in an embedding space is effectively eliminated, and cross-modal alignment errors are avoided from the source; by integrating the attention mechanism of modal perception in the graph neural network, the system can dynamically learn the semantic association strength between the nodes in different modals, adaptively adjust the weight distribution in the neighborhood information aggregation process, and significantly improve the accuracy of node characterization.

The invention discloses a knowledge graph enhanced reasoning method and a knowledge graph enhanced reasoning system for high-risk field medical decision, which are used for strictly limiting the reasoning of a large language model within a high-quality knowledge graph range, so that the reliability is improved, and factual errors and'model illusion 'are fundamentally eliminated. Through an innovative constraint generation and traceability mechanism, each conclusion can be traced to a direct evidence in the knowledge graph, the interpretability is enhanced, and the complete transparency of the reasoning process is realized. The mixed query driven sub-graph retrieval method can deeply understand the intention of a user, accurately position and extract structured knowledge related to complex problems, improves the retrieval precision, and has a retrieval effect far better than that of traditional keyword or vector retrieval.

The invention discloses a gear machine tool expert system construction method based on a generative rule, and the method comprises the steps: firstly carrying out the multi-dimensional modeling of a gear and a machine tool through an object-oriented technology, enabling a gear model to distinguish the characteristics of an axis position, a tooth curve and the like through an interface, and enabling a machine tool model to distinguish the machining technology and the precision grade through an interface; secondly, establishing a generation type rule mapping system from geometric, precision, material and functional attributes of the gear to machine tool movement, precision, process and rigidity attributes; finally, an intelligent reasoning system including forward reasoning, reverse verification, conflict coordination and self-learning iteration is constructed, and automatic and accurate matching from gear design requirements to machine tool optimization configuration is achieved. According to the method, the problems of high subjectivity and low efficiency caused by dependence on expert experience in the prior art are solved, and knowledge-driven intelligent decision making is realized.

A multi-hop reasoning method based on dynamic reasoning guidance and multistage self-feedback retrieval belongs to the field of natural language processing, and comprises the following steps: deconstructing a multi-hop reasoning process into a target-oriented sequence decision problem, carrying out dynamic reasoning guidance by using a large language model, generating a sub-problem sequence matched with a reasoning progress in real time, and carrying out multi-level self-feedback retrieval on the sub-problem sequence; target document retrieval is guided, and sub-questions are dynamically generated; according to the generated sub-questions, obtaining associated documents by adopting a three-level collaborative retrieval mechanism; and performing information refining on the associated document through a large language model, fusing the refined information into an inference chain, and performing inference to generate an answer. The invention further discloses a multi-hop reasoning system, a storage medium and a computer program product. The method aims at solving the complex multi-hop problem that multiple dispersed knowledge fragments need to be integrated, high-accuracy and high-efficiency reasoning is achieved, the retrieval requirement is dynamically generated through an explicit thinking chain guiding mechanism, and evidence obtaining is optimized and redundant information is filtered in combination with a three-level self-feedback retrieval mechanism.

The invention discloses a large model mixed load-oriented self-adaptive low-delay reasoning configuration generation method and device, computer equipment and a storage medium, and the method comprises the steps: determining a first token generation delay and an adjacent token delay interval which are historically configured on requests with different reasoning configurations, and obtaining tuples to form a configuration performance database, generating a delay prediction model by combining a least square method with the configuration performance database; dynamically dividing the historical request into a plurality of buckets according to the input length and the output length through a self-adaptive bucket dividing strategy; a configuration generator generates reasoning configuration for each bucket according to the input length and the output length of the historical request of each bucket; under the real mixed load, the problems of remarkable resource contention, queue head blockage, KV Cache switching overhead increase and the like are avoided in concurrent execution of long and short requests, and meanwhile, the problem of tail delay amplification is avoided, so that a reasoning system gives consideration to low delay and high throughput among different requests.

The invention relates to the technical field of robot intelligent control, and discloses an inspection robot task scheduling and state switching method for quantifying a state machine, which comprises the following steps: acquiring state machine operation configuration data; performing health degree evaluation on the sensor, navigation, communication and battery systems in an initial state; in the idle state, a segmented charging strategy is adopted, and task priorities are calculated through a neural network; a breakpoint resume and event-driven architecture is adopted to execute tasks in the inspection state; in a warehouse returning state, a deep neural network is adopted to predict return flight energy consumption, and a path is re-planned through multi-objective optimization when the electric quantity is insufficient; processing a control instruction by adopting speed limitation in a manual or mapping state, and starting an SLAM module in the mapping state; and in an abnormal state, a fuzzy logic reasoning system is adopted to calculate an abnormal grade, and recovery waiting, degradation protection or alarm is executed according to the grade. According to the invention, autonomous operation and intelligent decision making of the inspection robot in a complex dynamic environment can be realized.

A method may include: receiving a prompt and generated text from the LLM; computing an original token probability distribution for each token in the prompt and in the generated text; receiving a token position probability distribution for each token position in the generated text from the LLM; identifying keywords in the prompt; perturbing embedding vectors for the keywords used by the LLM by adding noise to the embedding vectors; computing a perturbed probability distribution for the perturbed embedding vectors by providing the perturbed embedding vectors as an input to a neural network used by the LLM, wherein the neural network returns a perturbed token probability distribution; evaluating a divergence between the original token probability distribution and the perturbed token probability distribution; identifying semantically meaningful tokens in the generated text; calculating a mean of divergences for the semantically meaningful tokens; and classifying the LLM based on the mean of divergences.

The invention discloses a collaborative reasoning method and system for adaptive model segmentation in a heterogeneous computing power environment. The method comprises the following steps: establishing a numerical calculation precision difference model between a source computing device and a target computing device; at a source equipment end, carrying out sensitivity analysis on a state tensor to be migrated according to the difference model, and asymmetrically dividing the state tensor into a core sensitive part and an edge part; performing dimension reduction processing on the edge part to generate a low-precision basic tensor, extracting precise residual information from the core part, compressing to generate a residual compensation vector packet, and sending the two to target equipment; and at a target equipment end, main calculation is carried out by using the low-precision basic tensor, a residual error compensation vector packet is decompressed in a bypass manner, and recovered residual error information is injected into an intermediate result of a main calculation stream to complete precision compensation. The problems of reasoning state semantic drift and error accumulation caused by dynamic migration between heterogeneous devices are solved, and the robustness and efficiency of a collaborative reasoning system are improved while the model reasoning accuracy is guaranteed.

The invention relates to the technical field of artificial intelligence, in particular to a target detection scene compliance reasoning system and method based on a multi-modal large model and GraphRAG. The method comprises the following steps: firstly inputting scene data into a multi-modal target detection and information extraction module, and outputting structured multi-modal detection information; the scene compliance rule is input into a scene rule knowledge graph construction module, and a searchable knowledge graph is output; the detection information and the knowledge graph are synchronously input into a GraphRAG retrieval reasoning module, and a reasoning result is output after entity matching, rule filtering and path sorting; a reasoning result is input into a compliance judgment and intelligent suggestion generation module, and a compliance conclusion and an intelligent suggestion are output; and when a new rule is accessed, dynamic updating of the knowledge graph is supported. The information extraction integrity is improved, accurate association of multi-modal detection information and compliance rules is ensured, cooperation of multi-modal target detection and structured compliance reasoning is realized, and the problems of detection information fragmentation, rule modeling non-structuring and opaque reasoning process in a traditional scheme are solved.

The invention relates to the technical field of cross of storage controllers and artificial intelligence acceleration, and discloses a large model reasoning system and method based on combination of a flash memory controller and an NPU (Network Processing Unit), and the large model reasoning system comprises the flash memory controller, the NPU and a flash memory array, the flash memory controller integrates a host interface module, a flash memory interface module, an independent AI acceleration interface module and an AI management engine, and the AI management engine autonomously completes NPU initialization, model weight direct loading, KV Cache hierarchical management, RAG knowledge base retrieval and model switching; the flash memory array is divided into a firmware partition, an AI special partition and a user storage partition, and different data storage requirements are met. According to the method, large model reasoning with low delay and low CPU dependence can be realized, and the model loading delay is reduced from 5-30 seconds to lt; after 500 milliseconds, the CPU occupancy rate of the host is reduced from 15-25% to lt; 2%, and concurrent operation of 4-8 models is supported. According to the invention, integration of storage and calculation is realized, and edge end, data center and mobile equipment scenes are adapted.

The invention belongs to the field of artificial intelligence, particularly relates to an interpretable artificial intelligence analysis and reasoning system based on image calculation, and aims to solve the problems that the AI image decision process is opaque and the causal logic is non-traceable. The system comprises a multi-scale feature extraction interface, a semantic concept generation interface, a causal inference engine interface, an anti-fact interpretation generation interface and a man-machine collaborative verification interface, and high-credibility image analysis is realized by constructing an interpretable path from pixels to semantic concepts and then to a causal chain.

This invention discloses a method and system for constructing an AI-native operating system based on ecosystem collaboration, belonging to the field of operating system technology. The invention first acquires and decomposes target requirements to form a set of functional and non-functional requirements, constraints, and priority lists. Based on this, it generates an operating system architecture description including kernel configuration and subsystem division. Then, it generates and completes multi-dimensional checks on module code and configuration files, and compiles, links, and packages them through a toolchain to generate an operating system image. This invention employs a three-layer AI model to realize requirement reasoning, system construction, and test optimization, forming a complete automated closed loop from requirement input to runnable image output. Simultaneously, it establishes a unified ecosystem collaboration mechanism to support multi-vendor collaborative development. This invention effectively reduces errors and repetitive work caused by manual intervention, shortens the development cycle, improves system stability and adaptability, and promotes the collaborative evolution and improvement of the AI-native operating system ecosystem.

The application discloses a low-altitude intelligent networking dynamic collaborative reasoning method based on multi-agent reinforcement learning, relates to the technical field of low-altitude intelligent networks and edge artificial intelligence, and comprises the following steps: preloading a light model and a complex model on each unmanned aerial vehicle (UAV), deploying a complete complex model by means of a ground station, and constructing an air-ground integrated intelligent reasoning system. By introducing an enhanced multi-agent deep reinforcement learning algorithm, each UAV can dynamically select a model type, determine a model segmentation point, and reasonably allocate bandwidth and ground computing resources based on the state of the UAV, network conditions and task characteristics during task execution, so that multi-DNN flow heterogeneous resource-aware collaborative reasoning is realized. The application can effectively improve reasoning accuracy and reduce average delay under different device performance, bandwidth conditions and task density, has good system scalability and adaptability, and solves the problems of limited single-machine processing capacity, restricted communication resources and inefficient model switching.

The application discloses an unmanned aerial vehicle intelligent sensing and intention reasoning system and device, which comprises image acquisition, target detection, knowledge graph, multi-modal large model, airborne computing power and edge-cloud cooperation module. The system is guided by target detection, enhances scene knowledge through the knowledge graph, improves the analysis ability of the multi-modal large model, and realizes the intelligent monitoring of the unmanned aerial vehicle based on the RK3588 platform. Real-time monitoring is realized by using the maneuverability of the unmanned aerial vehicle, which can not only analyze the scene, but also infer and predict the behavior intention and situation, thereby realizing early warning. The local and cloud systems are equipped with a historical database, a visual interface and a large parameter model, and the returned information is deeply analyzed. The system can be applied to the scenes of border patrol, city security, animal protection and the like.

The present disclosure generally relates to a framework for recommending and evaluating courses of action (COAs) using large language models (LLMs). In accordance with some aspects of the present disclosure, a system may generate a structured data representation using a retrieval augmented generation (RAG) system connected to data sources and an LLM. The system may provide the structured data representation to a graph-based logical induction with differentiable reasoning (GLIDR) system. The GLIDR system may select one or more solving units and provide data from the structured data representation to the selected solving units. The selected solving units may generate action data that causes an actuating device to perform one or more operations. The system may provide this action data to the actuating device. In some implementations, the structured data representation includes a knowledge graph or a graph schema.

The invention discloses an efficient edge reasoning system and method based on a lightweight multi-modal large model, and relates to the technical field of large model edge reasoning, and the method comprises the steps: S1, carrying out the time discretization and alignment segmentation of an original video frame, an audio signal and a text stream, and obtaining a lightweight feature vector, performing splicing and linear projection to obtain unified semantic feature vectors, and performing arrangement to obtain a unified semantic feature sequence; s2, performing memory reading based on the unified semantic feature vector and the memory state to obtain a memory reading feature, and performing calculation to obtain a surprise degree and a routing decision mark; s3, constructing a unified semantic feature sequence based on the routing decision mark and the short-term context window, and executing forward reasoning to obtain core semantic representation and a task output log value vector; and S4, constructing a write-in value vector based on the routing decision mark and the core semantic representation, performing surprise calculation on the write-in intensity, and obtaining the memory state of the next time step through low-rank outer product addition and two-norm truncation processing.

Pilot Cognitive Reasoning System and Method. A system for determining a pilot's cognitive state includes one or more processors coupled to a memory, the processors being configured to receive physiological data and aircraft status data; determine the pilot's mental workload and mental fatigue based on the physiological data; determine the pilot's available attentional resources based on the mental workload and mental fatigue; determine the pilot's attention allocation based on the available attentional resources and a gaze pattern derived from the physiological data; determine the pilot's situational awareness based on the attention allocation and the aircraft status data; and generate visualizations of the pilot's available attentional resources, the pilot's attention allocation, the pilot's situational awareness, and the aircraft status data.

The invention discloses an intelligent troubleshooting reasoning system and method for vehicle faults, and the system comprises a data preprocessing unit which converts a structured document and an unstructured document into structured data in a standard format; the diagnosis and repair knowledge graph unit is used for generating a diagnosis and repair knowledge graph from the structured data; the intelligent reasoning engine is used for providing recommendation operation according to the data of the vehicle terminal, the fault sensing information of the maintenance personnel and the diagnosis and repair knowledge graph of the diagnosis and repair knowledge graph unit, and sending the recommendation operation to the intelligent diagnosis center; and the intelligent diagnosis center is used for interaction between the user and the intelligent inference engine, provides storage, processing, transceiving of all data and services, provides a maintenance case management platform for the user, and processes and analyzes data and result data in the maintenance process. The invention relates to the technical field of vehicle detection and maintenance, and can solve the problems that in the prior art, manual vehicle fault troubleshooting and maintenance efficiency is low, and errors are prone to occurring.

The application discloses a breeding decision reasoning system for intelligent seed industry, relates to the technical fields of intelligent seed industry, edge computing and block chain, and is used for solving the technical problems of data security and sharing conflict of an existing breeding decision system, no independent reasoning capability of an edge terminal, and high dependence of a cloud terminal. The application adopts a block chain-edge computing deep cooperation architecture, comprises a perception layer, an edge computing layer, a block chain layer, a cloud terminal layer and an application layer; the perception layer collects multi-source breeding data, the edge computing layer deploys a lightweight Transform model to realize local real-time reasoning of field data, and can independently operate without the cloud terminal; the block chain layer realizes data security sharing and non-tamperable tracing through an improved consensus mechanism and privacy encryption technology; the cloud terminal layer completes model iteration and global scheduling, and the application layer provides a visual interactive interface. The application is suitable for field, greenhouse and remote breeding scenes, can significantly improve breeding decision efficiency and data security, and has good industrialization value.

The application discloses a kind of knowledge graph completion methods of combining relationship perception anchor point enhancement and graph convolution network, including in input data and construct relationship perception neighbor set, construct relationship perception anchor point layer, construct feature extraction layer, construct feature fusion layer, construct relationship reasoning layer, construct training layer and loss function layer.Affinity effect lies in: by introducing anchor point semantic information before knowledge reasoning, designing semantic-structure-anchor point three modal fusion mechanism and introducing anchor point consistency constraint term in loss function, so that generalization ability is superior and scalable, stable in zero-sample relationship and low-resource entity scene, modular design can be integrated into existing KGC or semantic reasoning system, can be widely applied in intelligent question answering, knowledge retrieval, recommendation system and medical knowledge reasoning etc.Scenario, with high precision, scalable, interpretable technical advantages.

The invention discloses a diagnosis and treatment data security sharing and collaborative reasoning system for a severe medical conjunct, and constructs a unified platform which integrates a standardized data interoperation layer, a security federated access control framework and an intelligent core engine for medical conjunct scene perception. Scattered medical institutions in the medical union are connected into a collaborative, efficient and intelligent severe illness treatment network, safe and real-time sharing of cross-institution data is achieved, intelligent agents are enabled to conduct collaborative reasoning, and therefore high-quality medical resources are sunk to each member unit in the medical union in an AI service mode, and the medical resources are saved. Knowledge and experience of top-level critical experts are converted into services which can be called anytime and anywhere, the problem that expert resources of primary hospitals are insufficient is effectively solved, and homogenization of the in-vivo critical diagnosis and treatment level of the medical combination is promoted.

The application relates to the technical field of large language models, in particular to a large language model streaming thinking and reasoning system and method, which comprises a streaming thinking data generation unit, a streaming training unit and a streaming parallel reasoning unit; the streaming thinking data generation unit is used for splitting continuous reasoning data into a plurality of reasoning thinking units to obtain a plurality of thinking segments; the streaming training unit adopts an independent position coding mechanism and a streaming mask mechanism to pre-train a large language model, so that the large language model is adapted to a streaming thinking paradigm; the streaming parallel reasoning unit uses the trained large language model to perform cross-GPU streaming parallel reasoning on the plurality of thinking segments, decouples an input token cache pool and a reasoning thinking token cache pool, and realizes cross-GPU concurrent execution of reading and thinking during reasoning. Compared with the prior art, the application has the advantages of significantly reducing delay and improving user interaction experience.

The application discloses a large language model end-cloud collaborative reasoning system based on low-rank fine-tuning, and belongs to the technical field of inference optimization of end-edge-cloud computing. An end-cloud collaborative reasoning architecture is established. In an offline stage, a large language model is fine-tuned based on training data of different downstream tasks on a cloud side. In an online stage, a user request is classified through a "variational autoencoder-Gaussian mixture model" clustering, and it is judged whether a low-rank adapter matching a current task exists in an end-side cache. If yes, reasoning is performed on the end side. Otherwise, the task is forwarded to the cloud side. After the architecture processes a plurality of user requests, a Mamba model is used to analyze user historical requests and a cache state, and an end-side low-rank adapter library is dynamically updated. End-cloud load and reasoning delay are monitored in real time, and a new adapter is incrementally issued to the end side according to a task repetition rate. The application realizes dynamic balance of the system, reduces computing and storage overhead, and ensures high efficiency and adaptability of the system.

The invention relates to the technical field of artificial intelligence acceleration, and discloses a DRAM-architecture-free large language model reasoning system and method based on an NAND flash memory, the DRAM-architecture-free large language model reasoning system is composed of a calculation unit array, a micro-cache, a flash memory interface controller, a storage arrangement engine and the NAND flash memory, a DRAM component is thoroughly omitted, and the calculation unit array is not influenced by the micro-cache. The model weight, the KV Cache and other core data are all stored in the NAND flash memory; an efficient direct connection data path is constructed through technologies such as calculation perception storage layout, KV Cache partition optimization, intelligent prefetching and self-adaptive error correction. According to the scheme, the hardware cost and power consumption are effectively reduced, the hardware structure is simplified, large-scale model off-line reasoning is supported, end side and edge calculation scenes are adapted, and reliability and practicability are achieved.

The application discloses a reasoning system and method which can be used for long text tasks, and the reasoning system comprises an information extraction and compression module and a reasoning module, wherein the information extraction and compression module is used for scanning background text, splicing an input query with the background text, appending a preset number of compression guide marks at the tail, compressing the spliced text, encoding the compression guide marks, and extracting the last layer of hidden states after coding as implicit fact marks; and the reasoning module is used for reasoning based on the implicit fact marks compressed by the information extraction and compression module. The reasoning system is used for reasoning, the knowledge extraction and the logical reasoning process are decoupled, the reasoning module can skip the original redundant text and directly perform efficient reasoning, and therefore, the reasoning delay is reduced while the effective context window is significantly expanded.

The invention discloses a multi-process task reasoning system and method and computer equipment. The multi-process task reasoning system comprises a scheduling process and a plurality of computing power nodes; the scheduling process responds to an external task reasoning request, analyzes the task and queries the current working state of each computing power node; the scheduling process selects the computing power node for executing task reasoning based on the analysis result of the task and the current working state of each computing power node; and the computing power node executes reasoning of the task and sends a reasoning result of the task to the scheduling process after execution is completed so as to update the processing state of the task. In this way, the flow sharing efficiency, the resource utilization rate and the stability caused by task reasoning single-process design can be improved.

The invention relates to the technical field of crossing of coal mine safety engineering and artificial intelligence, and provides a gas extraction and coal seam anti-reflection intelligent decision-making method based on a three-dimensional geological model, which comprises the following steps: extracting multi-dimensional geological parameters in a preset range of a working face to be mined and carrying out standardization processing to form a structured parameter matrix; inputting the structured parameter matrix into a pre-trained adaptive neural fuzzy inference system model, and outputting an extraction difficulty level score of the to-be-mined working face; according to the extraction difficulty level score, matching at least one gas extraction and coal seam anti-reflection technology combination from a preset structured measure library; and based on a mixed water circulation-moth fire optimization algorithm, engineering parameters in the matched technical combination are optimized, and an optimized gas extraction and coal seam anti-reflection construction scheme is output. According to the method, rapid and accurate optimization of multi-target engineering parameters is realized, and the scientificity and decision-making efficiency of a gas extraction scheme are remarkably improved.

The application relates to the field of information processing and discloses a knowledge management reasoning system based on accurate positioning retrieval, which is centered on the semantic anchor point, performs bidirectional logical correlation detection in the starting direction and the ending direction of the text respectively, extracts a first logical coupling degree when detecting in the starting direction, divides the preceding sentence into a slicing range when the first logical coupling degree is greater than a preset correlation threshold, extracts a second logical coupling degree when detecting in the ending direction, divides the subsequent sentence into the slicing range when the second logical coupling degree is greater than the preset correlation threshold, stops the detection until an adjacent semantic anchor point is extracted or the second logical coupling degree is lower than a preset breaking threshold, and generates a variable-length logical slice based on the sentence boundary divided into the slicing range. The application dynamically identifies a logical boundary through a semantic anchor point, eliminates the semantic breaking problem caused by mechanical cutting, and greatly improves the quality of knowledge management.

The application discloses an AI computing power energy efficiency improvement method and a distributed system based on software and hardware collaboration, and particularly relates to the technical field of artificial intelligence energy efficiency optimization, and is used for solving the problems of independent regulation and control of software and hardware layers, lack of collaborative optimization for dynamic load in the existing artificial intelligence reasoning system. Through the establishment of a closed-loop regulation chain of unified cycle observation, bottleneck migration determination, opportunity identification constraint, limited perturbation exploration, collaborative execution path generation and collaborative template updating under the reasoning session, the computing power mismatch, memory access mismatch, transmission mismatch and queuing mismatch in the model execution process are targeted and cooperatively adjusted in the online reasoning scene where long and short requests are mixed, load fluctuation and resource contention coexist, the unit reasoning energy consumption is reduced and the tail delay deterioration is inhibited, thereby solving the problem of unstable energy efficiency improvement caused by the isolated control of software and hardware layers in the prior art.