77 results about "Adaptive reasoning" patented technology

The invention discloses an adaptive question-answering system based on knowledge distillation and multi-modal dynamic fusion, and the system comprises a knowledge distillation module which is used for migrating knowledge of a teacher model pre-trained on corpora in the communication field to a lightweight student model, achieving model compression through optimizing a distillation loss function, and obtaining a multi-modal dynamic fusion model; the loss function comprises a soft label output by the teacher model and a KL divergence constraint output by the student model; the multi-modal knowledge fusion module comprises a feature extraction unit, a self-adaptive weighting unit and an attention fusion unit; the self-adaptive inference engine comprises a semantic analysis unit; according to the cross-modal reasoning method and system, semantic alignment of equipment parameters, protocol texts and topological graphs is achieved through the multi-modal dynamic fusion technology, and the cross-modal reasoning accuracy is improved; compared with an original model, the lightweight student model has the advantage that the reasoning speed is increased in a protocol analysis task.

The invention discloses an image generation method and system based on diffusion model adaptive reasoning. The method comprises the following steps: constructing a next noise intensity prediction module TPM; improving an original diffusion model, and adding a TPM; the improved diffusion model is trained by using a reinforcement learning method, the whole denoising process is expanded for multiple times, the next time step is predicted in each iteration, an award is calculated according to the quality of a finally generated image and the number of steps, and TPM parameters are updated by using a PPO loss function; in the actual reasoning process, noise is randomly sampled from Gaussian noise and input into the trained diffusion model, the diffusion model is used for predicting a denoising result, meanwhile, the TPM is used for outputting distribution of time steps based on next denoising, and a distribution mode is used as the next time step used in actual reasoning; and when the predicted time step is smaller than a set threshold value, a final image generation result is obtained based on model output. According to the method, the image generation process can be dynamically adjusted, and a reasoning result with both efficiency and quality is realized.

The invention relates to the technical field of large model application, and discloses a code generation task adaptive reasoning method, device and equipment. The method comprises the following steps: obtaining a problem description of a code generation task, and converting the problem description into a pseudo code based on a large language model; according to the problem description and the corresponding pseudo code, based on a pre-trained complexity evaluation model, performing complexity evaluation on the code generation task; the complexity evaluation model is obtained through training by taking problem description and pseudo codes as input and task complexity as output; a reasoning mode is matched according to the complexity evaluation result, a set reasoning model is adopted, and a code generation task is executed based on the matched reasoning mode; wherein different complexities correspond to reasoning modes of different reasoning depths. According to the method, the dynamic perception of the large language model on the task complexity is realized, so that the reasoning depth and the resource allocation can be adaptively adjusted according to the actual demand of the task.

The invention relates to a visual transformer substation intelligent inspection system and method based on a multi-mode large language model, and belongs to the technical field of power system intelligence. The system obtains image, temperature, vibration and noise data of substation equipment in real time through a multi-modal data acquisition module, and performs preprocessing and fusion. A high-precision three-dimensional semantic model is constructed by using a three-dimensional dynamic modeling module, and the device attributes are automatically labeled by fusing LLM semantic understanding capability. A multi-modal large language model (LLM) engine is combined with cross-modal feature extraction, a dynamic knowledge base and a self-adaptive reasoning unit to realize accurate diagnosis of equipment faults. An augmented reality (AR) interaction module displays the real-time state of equipment through AR glasses and supports natural language interaction. The self-interpretation decision support module generates interpretable fault reports and maintenance suggestions, and the communication and feedback module is responsible for data uploading and remote alarm. According to the method, multi-dimensional perception, dynamic knowledge reasoning and self-adaptive decision support of the equipment state are realized, the intelligent level of substation inspection is remarkably improved, the inspection efficiency is improved by more than 40%, the omission ratio is reduced to less than 1%, and rapid diagnosis of more than 95% of novel faults is supported.

The invention relates to the technical field of intelligent prediction and control technologies, in particular to an industrial hearth temperature prediction system based on a reasoning human learning optimization algorithm, and aims to solve the problems that in the prior art, temperature change cannot be accurately predicted under complex working conditions, adaptive adjustment cannot be performed under dynamic working conditions, and the prediction efficiency is low. And accurate prediction of a temperature field and combustion control optimization cannot be realized. The hearth temperature prediction model is constructed through the prediction model construction module to achieve high-precision prediction of the hearth temperature, through combination of Bayesian reasoning and the self-adaptive reasoning network, the system can accurately predict temperature changes under the complex working condition, response is fast, optimal combustion control is achieved, the system can conduct self-adaptive adjustment under the dynamic working condition, and the system is suitable for being used in the dynamic working condition. Accurate prediction and combustion control optimization of the temperature field are achieved, and the method has high robustness and is suitable for various high-temperature industrial scenes.

The invention discloses an air-ground dynamic collaborative reasoning method and device for a low-altitude Internet of Things. The method comprises the steps that a trained small lightweight deep neural network model and a trained large complex deep neural network model are deployed on a single unmanned aerial vehicle and a ground station respectively; the method comprises the following steps: taking an unmanned aerial vehicle as an air node, taking a ground station as a ground node, and after receiving a reasoning task request, carrying out statistical evaluation on own state and available resources by the unmanned aerial vehicle, and sensing the channel condition of air-ground transmission by the unmanned aerial vehicle; and selecting whether the unmanned aerial vehicle only uses the small lightweight deep neural network model to perform reasoning independently, if not, dividing the reasoning task into two sub-tasks according to the layer structure and the layer number of the large complex deep neural network model, and allocating the sub-tasks to the unmanned aerial vehicle and the ground station for common execution. According to the method, a flexible and efficient adaptive reasoning strategy can be provided for air-ground cooperative calculation, and the calculation efficiency of the low-altitude Internet of Things is greatly improved.

The invention relates to a large model adaptive reasoning method based on Monte Carlo tree search. The method comprises the following steps: organizing a reasoning process of a large language model by adopting a tree structure; selecting a node to be expanded by using a UCT algorithm; when a new node is expanded, all information on the current reasoning path is used for the currently selected node to generate a next reasoning step; simulation is executed, reasoning is continued based on the existing reasoning steps until a final answer is obtained, and whether the final answer is correct or not is judged; carrying out back propagation according to the correct final answer, carrying out reward updating on nodes on the current reasoning path, taking a tree structure before new nodes are generated as cue words, and taking the nodes generated according to the cue words as generated data; finely adjusting the large model according to the generated data; and carrying out reinforcement learning on the large model by using a PPO algorithm. According to the method, the father node can be automatically selected according to the current reasoning state, the new reasoning node is added into the nodes, and finally an accurate and reasonable answer is generated.

The invention provides a complex scene-oriented end-to-end semantic extraction system, belongs to the technical field of artificial intelligence and natural language processing, and realizes cross-modal information association through a multi-source heterogeneous data fusion module to construct a dynamic semantic network model. A hierarchical attention mechanism is adopted to carry out context-aware coding on unstructured input, and unsupervised pre-training and a weak supervised fine tuning strategy are combined to optimize a feature representation space. And designing an adaptive inference engine, automatically switching semantic analysis paths based on scene complexity, and generating a structured output result. According to the method, the dependency on specific knowledge in the field is reduced, the semantic understanding generalization ability in a complex scene is remarkably improved, high-precision analysis performance can still be kept in a low-resource environment, meanwhile, calculation resource consumption is reduced, and the method is suitable for practical application scenes with multi-language mixing, serious noise interference and high real-time performance requirements.

The invention relates to a basic model-oriented scene self-adaptive edge cloud collaborative reasoning system and method. The system comprises a scene self-adaptive edge side model customization component and a self-adaptive reasoning component; the scene-adaptive side model customization assembly extracts a hybrid expert model from a basic model deployed at a cloud side through a super-class routing and model distillation method, and then selects a proper expert module to compress and finely adjust a side customization model by using a small amount of label-free original data on edge equipment. And the adaptive reasoning component processes the reasoning task in cooperation with the side customization model and the basic model deployed on the cloud side. And uploading a reasoning result of the side customized model to a decision module, calculating a confidence score, and determining whether a data block selector needs to be uploaded or not, if so, uploading local and important data blocks to a basic model deployed at a cloud side, so that the effects of reducing transmission overhead and protecting data privacy are achieved.

The invention discloses a security access control method based on a zero-trust model, which comprises the following steps: collecting multi-source context information corresponding to an access request, and generating a context information set; performing fuzzy coding processing on the context information set to obtain a fuzzy context variable set; constructing an improved fuzzy Bayesian network model, and forming a context adaptive inference network; inputting the fuzzy context variable set into a context adaptive reasoning network, executing a fuzzy reasoning operation, and outputting an access risk level assessment result; generating an access control instruction through an access control decision mapping module based on the access risk level evaluation result; executing access response processing according to the access control instruction, and generating an access processing result; and jointly constructing strategy feedback information by using an access risk level evaluation result and an access processing result, and updating the context adaptive inference network. According to the invention, the improved fuzzy Bayesian network is adopted, and multi-context adaptive access control is realized.

The invention relates to the technical field of artificial intelligence, in particular to a gynecological traditional Chinese medicine data processing method and system based on four diagnosis data weight analysis. The method comprises the following steps: acquiring four diagnosis data; map mapping is carried out according to the four-diagnosis data to obtain four-diagnosis map mapping data; obtaining expert experience data and historical case data according to the four-diagnosis map mapping data; performing decision tree construction according to the expert experience data to obtain an expert experience judgment tree model; performing hierarchical weighted regression according to the historical case data to obtain historical case hierarchical weighted data; performing four-diagnosis feature weighting processing according to the expert experience judgment tree model and the historical case level weighting data to obtain four-diagnosis data weight analysis data. According to the method, accurate butt joint of four diagnosis features and a traditional Chinese medicine knowledge system is realized, and a complete link from four diagnosis original feature collection, semantic mapping and rule data collaborative modeling to context adaptive reasoning is constructed.

The invention discloses a method and system for enabling a large model to have spatial reasoning and long-term planning capabilities, and belongs to the technical field of artificial intelligence, and the method is implemented by the following steps: space-relation-graph conversion: converting complex spatial description into a graph structure which can be efficiently processed by a computer; adaptive reasoning: combining the language understanding ability of the large language model and the structure perception ability of the graph neural network to realize a reasoning mechanism; curriculum reinforcement learning: through curriculum reinforcement learning, a progressive learning strategy is used, so that the system can gradually master spatial reasoning tasks from simple to complex; by means of meta-cognitive monitoring, the system can evaluate the decision-making quality of the system in real time and adjust the decision-making quality when necessary; training and optimization: adopting an end-to-end training mode, and combining supervised learning and reinforcement learning to realize optimization. According to the method, the performance and robustness of the model in a complex space task are remarkably improved, and the adaptability and interpretability of the model are enhanced.

The invention relates to the technical field of intelligent building safety and artificial intelligence, and discloses an intelligent building dangerous case identification and grading early warning system based on multi-source data fusion, and the system comprises a data collection and management module which is used for obtaining and outputting a standardized multi-dimensional time sequence data flow and a building logic relation; the dynamic state modeling module is used for establishing a normal state model library for different space-time scenes; the event processing and analysis core module is used for calculating a risk entropy value representing an abnormal degree and a risk entropy gradient vector indicating an abnormal source; the adaptive reasoning and detecting module is used for performing source reasoning and risk diffusion prediction according to the risk entropy gradient vector; and the early warning and decision output module is used for judging the risk evolution situation and outputting a graded early warning signal. According to the method, the source of the dangerous case can be accurately positioned, the diffusion can be predicted, intelligent conversion from passive alarm to active detection is realized, and the accuracy and timeliness of early warning and the scientificity of decision making are remarkably improved.

The invention discloses a cross-modal-based multi-modal neural network dynamic reasoning path optimization method and system, and the method comprises the steps: carrying out the modal type classification of original multi-modal data, carrying out the quick feature extraction, carrying out the feature fusion of all modal early features obtained through the quick feature extraction, and obtaining multi-modal early features; reasoning path selection is carried out according to the multi-modal early-stage features to obtain each modal reasoning path, and feature fusion is carried out on the late-stage features of the modal obtained after full-quantity feature extraction is carried out on the modal selected as full-quantity feature extraction and the modal early-stage features corresponding to the modal selected as the modal skipping full-quantity feature extraction; and reasoning to obtain a specialized task result. According to the characteristic that a correct prediction result can be obtained by using a partial modal reasoning result according to some multi-modal tasks, an improved self-adaptive reasoning path selection method is constructed, the overall multi-modal network calculation amount is fully reduced, delay is reduced, energy overhead is reduced, and the prediction efficiency is improved. And finally, the deployment and landing of the multi-mode network in a future autonomous system scene are accelerated.

The invention provides a multi-model adaptive reasoning system and reasoning method, and is applied to the technical field of artificial intelligence. The system comprises four core modules: a user side is responsible for receiving and forwarding a request; the routing layer module serves as a decision-making center, analyzes request types and intelligently distributes the request types to the reasoning layer; the reasoning layer module executes a specific reasoning task and returns a result; the cache layer module stores a high-frequency request result to improve response efficiency; the implementation process of the method is as follows: submitting a request by a user, analyzing and distributing by a routing layer, processing and caching a result by a reasoning layer, and regularly optimizing a model based on cached data. According to the system, through a dynamic routing and intelligent caching mechanism, the consumption of computing resources is effectively reduced, and the response speed and the processing efficiency of the artificial intelligence service are improved.

The invention relates to an adaptive reasoning optimization method based on a target detection model in a limited environment, and belongs to the field of video target detection. According to the invention, under the environment that the CPU, the NPU and the memory resources are limited, the process of target detection processing and display of the video is completed, the video is disassembled into image frames, the image frames are distributed to a plurality of processes according to the frame sequence, and the processing of each frame of image is completed. Meanwhile, the invention provides a method for optimizing the total number n of processes in a limited environment, a method for calculating the FPS, a frame rate shaping output method and a method for setting the length of a multi-process processing queue. By optimizing process management, computing resource allocation and queue length setting, efficient and stable video target detection is realized on an embedded device or a mobile platform with limited resources.

The invention discloses an AI model dynamic optimization and adaptive reasoning method based on a client, and relates to the technical field of model optimization, hardware resources and an operation environment of client equipment are analyzed, then a proper AI model is selected and deployed to the client equipment, and then operation state information of the client equipment is monitored in real time. Therefore, the AI model is dynamically optimized and adjusted to achieve the purpose of efficient reasoning, an adaptive algorithm is obtained through analysis according to the hardware resource information of the client equipment, the resource occupation amount of each reasoning path and the reasoning accuracy rate, the client equipment can adjust the optimal reasoning path in real time according to the adaptive algorithm, and the reasoning efficiency is improved. Therefore, the optimal reasoning performance and efficiency are realized under the condition of limited resources, and finally, each processed task and the corresponding AI model and the optimal reasoning path are stored to the cloud for subsequent direct calling.

A computer-implemented method, according to one approach, includes: receiving, at a context-aware multi-model aggregator, a user query from an endpoint device. The user query is evaluated, and models and / or combinations of models are selected to evaluate the user query based at least in part on the evaluation of the user query. An adaptive reasoner of the context-aware multi-model aggregator is used to select an output to the user query based at least in part on results of the selected models and / or combinations of models. The output is transmitted to the endpoint device, and reinforcement learning is performed based at least in part on feedback received from the endpoint device.

The invention discloses a real-time dynamic reasoning optimization method and system for an AI gateway, belongs to the technical field of artificial intelligence and edge computing, and aims to solve the technical problems of how to optimize the real-time dynamic reasoning capability of the AI gateway, improve the real-time reasoning efficiency of the AI gateway under complex tasks and improve the real-time reasoning efficiency of the AI gateway under complex tasks. According to the technical scheme, the method comprises the following steps: model lightweight preprocessing: constructing an AI model by adopting a lightweight network architecture, and reducing the model calculation complexity and the number of parameters by using a deep separable convolution and channel shuffling technology; carrying out model compression processing of pruning, quantification and knowledge distillation on the constructed model; real-time dynamic reasoning and decision making: in the operation process of the AI gateway, monitoring the complexity of input data, the real-time requirement of a current task and the use condition of computing power resources of the AI gateway in real time, and performing self-adaptive reasoning strategy decision making according to a monitoring result; reasoning engine and compiling optimization; data preprocessing and edge intelligence are realized; and edge-cloud collaborative reasoning.

The invention provides a nuclear power plant anomaly detection method and system based on auto-encoder edge calculation, and belongs to the technical field of nuclear power plant anomaly detection. Comprising the following steps: establishing multi-dimensional time sequence data by using key operation parameters of nuclear power plant equipment, and preprocessing the multi-dimensional time sequence data into a time window sample; calculating a reconstruction error according to the time window sample by using an auto-encoder and generating an abnormal score; performing anomaly judgment according to the reconstruction error and the anomaly score by adopting a threshold self-adaptive reasoning algorithm, and feeding back an anomaly judgment result to the central server; and updating the model parameters of the auto-encoder in combination with incremental learning, and issuing the model parameters to the edge nodes through a streaming updating mechanism to realize updating of the auto-encoder. According to the invention, continuous monitoring and adaptive optimization of the operation state of the nuclear power plant equipment can be realized.

The invention relates to the technical field of artificial intelligence and natural language processing, and discloses a large language model reasoning enhancement method based on multilayer reasoning chain verification, which comprises the following steps: constructing a multilayer reasoning architecture based on a pre-trained large language model, and establishing a reasoning state space and a reasoning step transfer function; receiving an input question, generating a parallel inference chain according to the multilayer inference architecture, and determining an inference step sequence and a confidence score of the parallel inference chain; performing consistency score calculation and logic consistency check according to the parallel reasoning chain, and determining a final answer according to the consistency score and the confidence score; and calculating complexity according to the input problem so as to perform adaptive reasoning depth adjustment, determining a reasoning process according to the reasoning depth, and outputting a complete reasoning result. According to the method, a multi-layer parallel reasoning architecture is constructed, and a reasoning chain cross validation mechanism and a self-adaptive reasoning depth adjustment strategy are combined, so that precise processing of a complex reasoning task is realized.

The invention provides a document structuring task processing method and system and a readable storage medium. The method comprises the steps of firstly obtaining a structured task request and corresponding unstructured document data, then dynamically determining an adaptive reasoning mode in combination with a task demand, a current time period and a load state of a preset processing model group, and associating the two modes with an exclusive target reasoning model in the model group respectively. In a high-load period or when a task needs to be quickly responded, a result-oriented reasoning mode is started, a structured document is directly generated by a correlation model, the process is simplified, and resource consumption is reduced; and when a load period or a task needs a clear basis, switching to an interpretable reasoning mode, synchronously outputting a structured document and a complete reasoning process by the association model, and clearly displaying a processing logic. According to the dynamic adaptation mechanism, the real-time performance of a high-demand scene is guaranteed, the interpretability defect of a model is overcome, and the real-time performance and decision transparency of text structured processing are doubly improved.

The present application discloses a method for dynamically optimizing an AI model and adaptive inference based on a client, which relates to the technical field of model optimization. By analyzing the hardware resources and operating environment of the user-side device, a suitable AI model is selected and deployed to the user-side device, and then the operating status information of the user-side device is monitored in real time, so as to dynamically optimize and adjust the AI model to achieve the purpose of efficient inference. At the same time, according to the hardware resource information of the user-side device, the resource occupancy of each inference path and the inference accuracy, an adaptive algorithm is analyzed. According to this adaptive algorithm, the user-side device can be enabled to adjust the optimal inference path in real time, thereby ensuring the best inference performance and efficiency under limited resource conditions. Finally, each processed task and its corresponding AI model and optimal inference path are stored in the cloud for subsequent direct invocation.

Embodiments provide an apparatus of a wireless communication network, the wireless communication network using one or more Artificial Intelligence / Machine Learning, AI / ML, models for one or more use cases, wherein the apparatus is to determine an inference time for one or more of the AI / ML models to be used in one or more network entities of the wireless communication network.

The invention discloses a dynamic tactical knowledge graph construction and target attackable value analysis system and method based on an environment embedded graph neural network, and belongs to the technical field of military intelligent decision making and firepower command control. The system constructs a tactical knowledge graph fused with firepower application expert experience for tactical rule recessive and environmental constraint dynamic problems in target threat assessment. Based on an entity-relation-environment attribute triple structure, an ontology model covering a typical combat scene is established, and dynamic quantitative analysis of target attackable value is supported. An environment embedded graph neural network is innovatively designed, a battlefield environment is used as an internal structure variable of a graph, and deep coupling and self-adaptive reasoning of knowledge representation and a real-time situation are achieved. The system can comprehensively consider information such as enemy conditions, my conditions, typical scenes and the like, realizes accurate evaluation of multi-target threats and values, assists a commander in accurately capturing important high-threat high-value targets, and quickly forms a wave-order striking target list.

The application relates to the technical field of laser system control, and discloses a multi-stage PID parameter regulation and control method for a laser frequency locking system based on deep learning, which comprises the following steps: acquiring an error signal time sequence of a PDH laser frequency locking system; outputting a frequency locking state score and fast-slow double-loop PID parameters through a deep learning model; judging whether the frequency locking state score and the fast-slow double-loop PID parameters both satisfy corresponding preset conditions; if the frequency locking state score and the fast-slow double-loop PID parameters do not both satisfy the corresponding preset conditions, iteratively optimizing the deep learning model until the frequency locking state score and the fast-slow double-loop PID parameters both satisfy the corresponding preset conditions; and outputting the fast-slow double-loop PID parameters of the optimized deep learning model. The application can realize online self-adaptive reasoning and closed-loop optimization of fast-slow double-loop multi-stage PID control parameters, and improve the intelligent level and the ability to adapt to complex working conditions of the PDH laser frequency locking system.

The present application belongs to the technical field of rotating machinery state monitoring and industrial deployment, and particularly relates to a reducer input shaft end-to-end rotating speed adaptive inference method based on variable-length sequence. In view of the fact that actual monitoring data is a continuous sequence of any length, and a deep learning model usually relies on fixed-length input, and traditional sliding window inference is prone to introducing phase discontinuity and splicing artifacts at the boundary, the present application establishes a deployment paradigm of "fixed-length training and variable-length inference". The continuous variable-length sequence is divided by a sliding window with an overlap rate, and a backtracking interception strategy is designed for the case of insufficient end sequence. After batch inference using a pre-trained model, the local prediction results are integrated and smoothed by using an overlap addition mechanism and a weighted average algorithm. The present application effectively eliminates the Gibbs effect at the window boundary, ensures the phase integrity of the terminal, and realizes seamless and smooth rotating speed curve reconstruction of continuous monitoring data of any length.

The invention discloses an electric power intelligent number asking method fusing adaptive reasoning and SQL (Structured Query Language) rectification, which relates to the technical field of electric power system intelligent number asking and comprises the following steps: A, converting a natural language query request proposed by a user into an SQL query expression conforming to an electric power business scene; b, introducing an SQL correction scheme based on abstract AST, performing structured analysis on the SQL query expression, identifying logic errors existing in SQL grammar and semantics, and performing correction and optimization; and C, the big language model carries out structured summarization on the original data obtained by query, and feeds back the data to a user in a natural language form. According to the method, the defects in the prior art can be overcome, and the data query efficiency and decision support capability of the power regulation and control business are improved.

The invention provides a self-adaptive reasoning visual question answering system and method based on humanity and geography common sense enhancement, and relates to the technical field of artificial intelligence, and the system comprises an obtaining module which is used for obtaining an input image and an input question; the image content and context analysis module is used for carrying out image content and context analysis on the input image and the input question to obtain context representation; the humanistic geography common sense retrieval and multi-hop reasoning module is used for performing problem decomposition on the context representation to generate a plurality of sub-problems, performing humanistic geography common sense retrieval and multi-hop reasoning on each sub-problem, and outputting a reasoning chain containing sub-problem-clue pairs; and the answer generation module is used for decoding the input question and the reasoning chain to obtain an answer of the input question. According to the method, the problems of low reasoning efficiency and insufficient accuracy of the existing humanity geography visual question-answering method are effectively solved.

The application discloses a kind of multi-modal psychological intention understanding methods based on zero sample and intention perception alignment, it is related to mental health monitoring and intelligent dialogue system technical field, including the following steps: step one, acquisition contains text, audio and video time synchronization dialogue data, constitute multi-modal data;Step two, construct multi-target fusion model, obtain multi-target joint optimization function by the multi-modal data of time sequence alignment combined with late fusion strategy;Step three, by adaptive reasoning strategy, multi-modal data is distributed according to dynamic calculation, and the predicted intention classification result is obtained.The application has the characteristics that it can be started and run without relying on large-scale artificial labeling data, fundamentally solves the bottleneck of high data labeling cost and difficult acquisition in the field of psychological dialogue, realizes accurate fusion of cross-modal information through time sequence alignment and significantly improves the accuracy of analysis.