156 results about "Knowledge learning" patented technology

The invention discloses an intelligent question-answering system training method based on machine learning, and the method comprises the steps: optimizing a knowledge learning sample to obtain optimized knowledge, determining a knowledge weight, constructing an industry knowledge base, fusing multi-modal question content to obtain fused question content, and carrying out the training of the intelligent question-answering system. Obtaining industry description information and a user portrait according to the fused question content and the user information, and inputting the fused question content, the industry description information and the user portrait into a question and answer reinforcement learning model to obtain answer content, and obtaining a first reward signal, a second reward signal and a dynamic reward weight according to the answer content and the user question response information, adjusting the question and answer reinforcement learning model, inputting the multi-modal question content into the adjusted question and answer reinforcement learning model to obtain optimized answer content, and updating the industry knowledge base by adopting a user interaction log and the knowledge weight. According to the method, intelligent question and answer system training can be accurately and efficiently carried out, and meanwhile powerful support is provided for intelligent question and answer services of all industries.

The invention provides a distillation method and device based on AI large model knowledge and electronic equipment, and belongs to the technical field of artificial intelligence, in the method, a dynamic adjustment mechanism is adopted for the temperature, and along with the increase of a training period, the preset initial temperature corresponding to the training period is gradually reduced; besides, in each training period, the preset initial temperature is finely adjusted based on the distillation loss, so that the target temperature corresponding to each training sample is obtained, namely the temperature most suitable for the current distillation effect. The target distillation loss and the student loss obtained by recalculating based on the target temperature are more scientific, so that the total loss for updating the weight parameters of the student model is more accurate and robust, the student model is prompted to learn the knowledge of the teacher model more effectively, and a scientific and efficient learning process is learned. The finally obtained student model has good generalization ability, and the robustness of the model is significantly enhanced, so that the overall effect of knowledge distillation is significantly improved.

The invention provides a robot social adaptive navigation knowledge learning and migration method and system, and relates to the field of mobile robot navigation. Aiming at the problems that an existing path planner lacks time sequence memory and neglects pedestrian social intent, a man-machine co-fusion scene is constructed, a training set containing an expert demonstration path is made, and a recursive generation model is input; designing a recurrent neural network embedded RRT, generating an RNN-RRT planner, and fusing historical information and pedestrian convergence probability in training; new scene loading training parameters are finely adjusted to realize knowledge migration, loss convergence or output RNN final parameters after reaching a preset round number. According to the method, the path anthropomorphism and generalization ability are improved, and the method is suitable for complex human-computer interaction scenes.

The invention discloses a software process arrangement system and method based on a large model, and particularly relates to the technical field of software process arrangement. The intelligent process design module is used for realizing intelligent and low-threshold process design and optimization through natural language understanding and a visualization technology; the Agent collaborative execution module is used for dynamically distributing tasks, monitoring states and coordinating a plurality of Agents to efficiently complete process execution; and the knowledge learning enhancement module is used for constructing a domain knowledge system and providing intelligent support for process decision and optimization. According to the method, the demand described by a natural language of a user is converted into a structured process definition through the intelligent process design module, rapid design is carried out by utilizing a process pattern library and a visual editing tool, and meanwhile, compliance check and optimization suggestion are carried out, so that compliance and high efficiency of process design are ensured, and the process design efficiency is improved. And the Agent collaborative execution module dynamically allocates tasks to appropriate Agents according to process requirements, monitors the execution state in real time, and performs coordinated processing in case of abnormality to ensure efficient execution of the process.

The invention discloses an enterprise innovation risk assessment system based on virtual-real fusion common knowledge learning, and the system comprises a data collection and preprocessing module which is used for collecting and preprocessing a real operation standardized data set; the simulation environment construction and disturbance generation module is used for constructing a virtual simulation environment and generating a virtual data set; the semantic alignment and fusion module is used for performing semantic alignment and weighted fusion to generate a virtual-real fusion data set; the risk relation graph construction module is used for constructing an enterprise innovation risk relation graph; the multi-order structure embedding module is used for executing multi-order graph structure embedding; the common knowledge migration module is used for extracting a common feature structure and executing migration adaptation; and the risk assessment module is used for carrying out risk assessment. According to the method, enterprise atlas and multi-semantic structure embedding are fused, cross-enterprise risk migration assessment is achieved, and the method has the advantages of being high in expression ability, high in adaptability and accurate in assessment.

The invention relates to the technical field of cross-border finance, and discloses a cross-border finance compliance examination intelligent engine based on a large model, which comprises a data acquisition layer, a data preprocessing layer, a large model core processing layer, a compliance examination decision-making layer and a user interaction layer. According to the cross-border financial compliance inspection intelligent engine based on the large model, the strong natural language processing ability and knowledge learning ability of the large model are utilized, unstructured text information can be effectively processed, fast learning is achieved, the engine adapts to constantly changing supervision policies, the large model further has the autonomous reasoning and analysis ability, and the large model can be widely used. Complex cross-border financial businesses can be comprehensively judged, and the defects of the traditional technology in the aspects of examination efficiency, accuracy and adaptability are overcome.

The invention discloses a knowledge learning management method and device based on data analysis, and the method comprises the steps: obtaining a scene behavior data set, the scene behavior data set comprises the behavior data of each pharmacy employee in different business scenes, and the business scenes at least comprise a drug recommendation scene, a drug taboo reminding scene and a special medication consultation scene; constructing a behavior model according to the scenarized behavior data set, and extracting behavior feature information of the drugstore employees in different business scenes through the behavior model; determining a knowledge point set and a knowledge blind area which need to be mastered by the drugstore employee in each business scene based on the behavior feature information; constructing a knowledge graph according to the scenarized behavior data set, the knowledge blind area, the knowledge point set and the association relationship between the knowledge point set and the business scene; and based on the knowledge graph, determining professional learning contents of the drugstore employees and priorities corresponding to the professional learning contents. The professional knowledge level of drugstore employees can be comprehensively improved.

The invention relates to the technical field of large-scale language models, and discloses a method for converting a natural language to an SQL (Structured Query Language) task by combining fine tuning and Prompt. According to the method, the database mode knowledge learning module is constructed, so that the database mode can be efficiently understood and memorized, the length of input tokens is remarkably reduced, and the token efficiency is improved. Meanwhile, the module utilizes a LoRA model to carry out fine adjustment on a large language model, the performance of the model on text-to-SQL tasks is optimized, and the model can generate SQL statements more accurately; besides, an example selection module is also constructed, and the most similar example is selected as a prompt by calculating the similarity between the problem and the candidate example, so that the accuracy of generating the SQL statement by the model is further improved. According to the method, the generalization ability of the model is improved, and the model can more flexibly cope with natural language queries in different fields.

The invention provides an intelligent curriculum recommendation method and device based on a knowledge graph and a storage medium, and the method comprises the steps: constructing a multi-dimensional knowledge graph which comprises a curriculum skeleton knowledge sub-graph, a user capability sub-graph and a scene demand sub-graph; generating a knowledge dependence matrix based on the course skeleton knowledge sub-graph, and generating user learning sequence features based on historical learning data of the user; acquiring a learning target, a learning time constraint and a learning scene input by a user, and determining a first knowledge learning path and task sequence set and a second knowledge learning path and task sequence set of the user; and based on the user learning sequence features, the course skeleton knowledge sub-atlas, the user capability sub-atlas and the scene demand sub-atlas, using a fusion model to perform fusion processing on the first and second knowledge learning paths and the task sequence set to obtain a course meeting the user demand. The accuracy of course recommendation is improved.

The invention discloses a voice and music collaborative generation method and system based on a dynamic mixed attention and expert architecture, terminal equipment and a medium, and relates to the technical field of audio generation. The method comprises the following steps: acquiring multi-modal input containing at least one of audio, text and vision, performing embedding processing on the multi-modal input, and mapping the multi-modal input to a unified space to obtain a fusion sequence; setting a plurality of attention heads, dynamically selecting the attention heads to activate experts through attention head routing gating, and weighting and aggregating calculation results to obtain mixed attention features; generating expert probability distribution through expert set routing gating, dynamically selecting an expert set to activate experts, and calculating task adaptation features; and performing audio generation processing based on the audio language modeling head to obtain a voice or music result. According to the method, collaborative generation of voice and music, dynamic allocation of computing resources, balance of domain-specific and cross-domain general knowledge learning are realized, the problems of task conflict and data imbalance are solved, and the audio generation quality and efficiency are improved.

The invention belongs to the technical field of electric power artificial intelligence, and discloses an electric power knowledge question-answering method based on a knowledge reasoning thinking chain embedded large model and a related device. The method comprises the steps of obtaining a power problem text; and inputting the power question text into a pre-trained knowledge reasoning thinking chain embedding large model, and outputting a text answer of the power question. According to the method, a supervised learning and preference optimization mechanism is fused, and collaborative optimization of large-model professional knowledge learning and thinking path modeling is realized by constructing an electric power knowledge question and answer and analyzing a thinking chain data set; the technical problems that an existing large model is weak in knowledge, inconsistent in thinking process and insufficient in generalization ability in power professional text questions and answers are solved.

The invention belongs to the field of sugar detection, and particularly relates to a litchi sugar detection method based on machine learning, and the method comprises the steps of complete information collection, priori knowledge learning, feature fusion, litchi model updating, litchi reflection spectrum scanning and recording, preprocessing, interpretability analysis and model output. According to the scheme, litchis automatically roll before passing through a detection area by adopting the rotating roller, and cameras are respectively mounted at different angles of the front and rear sections of the rotating roller and the left and right sides of the conveying belt, so that errors caused by single-angle shooting are reduced, and the sugar detection accuracy is improved; according to the scheme, a nondestructive, rapid and low-cost near-infrared spectral imaging technology is adopted, near-infrared wavelengths are screened according to the importance of the wavelength to a sugar detection result, redundant information is removed, features contributing to detection and grading are extracted, an interpretable artificial intelligence technology is adopted, and the design of a return model is optimized.

The invention is suitable for the technical field of education, and provides a content recommendation method and system based on children cognition, and the method comprises the steps: obtaining historical learning data, carrying out the construction of a knowledge graph based on the historical learning data, and generating a knowledge structure graph; performing vector mapping processing on the plurality of knowledge nodes and the edges connecting the plurality of knowledge nodes by using a graph embedding algorithm to generate a knowledge vector set; according to the knowledge vector set, calculating a semantic association degree among the plurality of knowledge nodes, and if the semantic association degree is greater than a preset threshold value, extracting a key knowledge node from the plurality of knowledge nodes by using a hierarchical clustering algorithm; determining a learning sequence corresponding to the key knowledge nodes according to a preset knowledge learning strategy, and generating a knowledge progressive sequence; generating a target knowledge node sequence based on the knowledge progressive sequence and the historical learning data by using a preset reinforcement learning algorithm; and the target recommendation content is generated according to the target knowledge node sequence and recommended to the user, so that more appropriate learning content can be recommended.

The invention discloses a robot manipulator grabbing and docking method and device, a medium and a product, and relates to the field of robot and sensor perception.The method comprises the steps that a visual-tactile joint data set of a manipulator in the process of grabbing a filter hose is constructed; establishing a tactile priori knowledge learning model according to the visual-tactile combined data set; outputting a current grabbing configuration according to the tactile priori knowledge learning model; motion planning information of the manipulator is obtained according to the current grabbing configuration; the manipulator is controlled to be in butt joint with the filter hose according to the motion planning information of the manipulator, mechanical modeling is conducted on the contact state of a butt joint of the filter hose and a butt joint of a machine tool, and a mechanical control model is obtained; and according to the mechanical control model, a DDPG parameter identification algorithm is adopted, and pose information and contact force information in the mechanical arm butt joint process are obtained. According to the flexible pipe grabbing and butting device, grabbing and butting of the flexible pipe of the cutting fluid filtering device can be more efficiently and stably carried out.

The invention provides a highway service area traffic flow scheduling method and system based on an intelligent algorithm, and the method effectively utilizes a to-be-learned template set comprising a plurality of template scheduling learning data through a deep knowledge learning algorithm. According to the method, in a target training stage, aiming at each template scheduling learning data, an effective labeling training value and an ineffective labeling global training value of a corresponding training traffic flow scheduling network relative to a traffic flow scheduling network to be called are dynamically evaluated, so that the pertinence and the high efficiency of a training process are ensured. And through continuous optimization learning, when the training error does not continue to decrease any more, deep knowledge learning is terminated, and finally, a target traffic flow scheduling network capable of accurately determining target traffic flow dynamic data scheduling knowledge description is obtained. According to the method, the intelligent level of traffic flow scheduling is improved, the traffic flow scheduling effect of the highway service area is optimized, the traffic jam of the service area is relieved, and the operation efficiency and the user experience of the service area are improved.

The invention discloses a deep learning model-based knowledge learning agent training method, and relates to the technical field of auditing, and the method comprises the steps: carrying out the optimal disassembly task classification through a multi-model fusion knowledge classification method, and obtaining defined auditing knowledge and undefined auditing knowledge; optimal disassembly task classification is carried out by using a multi-model fusion knowledge classification method, and undefined audit knowledge is analyzed through an adaptive improved iteration mechanism, a fusion NLP algorithm and a deep learning audit agent to generate a question answer; the data security in the use process of the audit knowledge learning agent is controlled through a dynamic authority control and question and answer tracing mechanism; when facing large-scale and multi-level auditing tasks, the system can quickly position the core of a problem, redundant operation is reduced, and the response speed is increased.

The invention discloses a two-stage knowledge distillation method based on dynamic sampling prompt. The two-stage knowledge distillation method comprises the following steps of image encoder distillation and decoder distillation. Based on the design of two-stage knowledge distillation, compression and light weight of a large model are realized, synchronous distillation is carried out on an SAM image encoder and a decoder, efficient alignment of encoder extraction features and a mask decoder is realized, and the calculation cost is effectively reduced while the excellent generalization ability of the model is reserved. According to the method, a dynamic sampling prompt strategy is designed in decoder knowledge distillation and is used for efficiently selecting prompt points, so that the model learning process is optimized. The strategy particularly focuses on incorrect regions in the segmentation result to guide the student model to be closer to a true value. By dynamically updating the prompt point set, knowledge learning of the student model to the teacher model is gradually enhanced, and performance loss possibly caused in the distillation process is effectively reduced.

The invention discloses a big language model knowledge editing system based on mixed experts, and belongs to the field of natural language processing. According to the method, the hybrid expert architecture and the keyword attention router are combined, so that the model knowledge is dynamically updated under the condition that the original parameters of the large language model are kept unchanged; according to the single-layer bypass hybrid expert adapter provided by the invention, only single-layer additional experts are introduced into a model, and input with similar knowledge requirements is routed to the same expert through a keyword attention router, so that the experts can efficiently distinguish and process different types of knowledge information; the invention further provides a data batch processing method based on semantics, similar instances are grouped in the training stage, specialization of an expert model is promoted, and knowledge learning preference of a large language model is better met. According to the method, excellent performance is shown on models of various types and scales and in various editing tasks, and balance between generalization ability and local optimization is achieved.

The invention belongs to the technical field of intelligent education, and provides a subject knowledge learning method planning system, which comprises an input data acquisition module, a knowledge graph analysis module, a knowledge graph planning classification module, a content classification module, an association combination output module, an abstract concept learning path module, an evaluation module and a planning module. Wherein functions of the output module and the planning module are associated and combined through the content classification module. And the content classification module performs third-level classification on the data subjected to second-level classification to obtain hierarchical abstract concept data information. And the association combination output module outputs learning content and abstract result data by using a neural network model, and performs grading according to difficulty. And finally, a planning module plans learning methods of abstracted and non-abstracted contents of the current subject for the user step by step according to the answering accuracy of the evaluation module.

The invention relates to a large language model-based job matching scene query understanding method. The method comprises the following steps of: receiving a job name, a job description and job requirement information input by a user; performing noise filtering and coding normalization processing on the input post information to generate a standardized text; geography and organization entities are extracted through a named entity recognition model, cue words are finely adjusted through a large language model, structural constraint conditions are formed by technical recognition educational background, working years and ages, and a condition analysis template is dynamically generated. And finally, through a large language model, verifying the accuracy of the keyword items, and allocating weight coefficients for the lexical items. Therefore, the efficiency is remarkably improved. Through automatic semantic analysis of a large language model, manual deep intervention of professional knowledge learning of posts is not needed, and core information such as post responsibilities and skill requirements can be rapidly extracted and converted into semantic retrieval features.

The invention relates to the technical field of data processing, in particular to a model training method and device, equipment and a storage medium. The method comprises the following steps: acquiring first input data and first output data of a first model; inputting the first input data as second sample input data into a second model, and obtaining second output data output by the second model; obtaining a first loss value based on the first output data and the second output data, and obtaining a second loss value based on the second output data and the target answer data; and training the second model based on the weighted sum of the first loss value and the second loss value to obtain a trained target second model. According to the invention, one-sided knowledge learning of the second model in the process of distilling knowledge from the first model to the second model can be avoided, the influence of error noise of the first model is avoided, and the training effect of the second model is improved.

The invention discloses a camouflage target segmentation method based on arbitrary modal assistance. The detection precision of a camouflage target is improved through multi-modal data fusion. According to the method, a multi-modal divider UniSEG and a cross-modal knowledge learning network UniLearner are included. The UniSEG adopts a double-branch architecture, extracts the features of RGB images and other modal images, and carries out preliminary fusion through a potential space fusion module LSFM. The SSFM is combined with a cross-state space model (CSSM) to further fuse features in a unified state space. And the UniLearner learns a mapping relation between the RGB image and the target mode through an encoder-decoder structure, generates a pseudo-modal image and a semantic-rich potential vector, and injects the pseudo-modal image and the semantic-rich potential vector into a specific layer of the UniSEG to improve feature extraction and fusion effects. The method has plug-and-play flexibility, can be seamlessly integrated into an existing segmentation network, is widely applicable to the fields of ecology, medicine, surface monitoring and the like, and remarkably improves the performance and robustness of disguise target segmentation.

The invention relates to the technical field of simulation practical training, in particular to a scene perspective simulation platform and a practical training method thereof, and the simulation platform comprises a theory and model simulation module and a practical training operation training module. The theory and model simulation module comprises a theoretical knowledge learning sub-module for outputting theoretical knowledge points; the model operation sub-module is used for displaying and executing splitting, combining and restoring operations of an operation user on the three-dimensional equipment model; the simulation analysis sub-module is used for identifying and recording operation behaviors of the operation user; the practical training operation training module comprises a practical training knowledge learning sub-module used for outputting practical training knowledge points; the practical training operation sub-module is used for realizing state transition of the three-dimensional equipment model according to the first practical training instruction; and the practical training recording sub-module is used for recording all first practical training instructions and operation results of the operation user. Through the theory and model simulation module and the practical operation training module, theory teaching and practical operation training can be effectively linked.

The utility model discloses a physical intelligence comprehensive training ground, which relates to the technical field of physical intelligence comprehensive training, and comprises a field and a plurality of letter-shaped and / or digital-shaped training frame bodies, and the plurality of training frame bodies are uniformly arranged in the field; the selected marking part comprises a selected switch and a display component used for marking and displaying selected signals of the selected switch; the number of the selected switches is multiple, and the multiple selected switches are arranged in one-to-one correspondence with the multiple training frame bodies, so that when each selected switch is turned on, a selected signal indicating that the corresponding training frame body is selected can be transmitted to the display assembly; the display assembly can receive the selected signal and mark and display selected information of the corresponding training frame body. According to the intelligent physical training device, people can be intelligently trained while carrying out physical training, and certain knowledge learning tasks can be completed while physical training is carried out.

The invention discloses a knowledge exchange method based on learning before forgetting and application thereof, and relates to the technical field of machine learning and knowledge management, a pre-trained deep learning model is selected as a basic model, and a retention set, a forgetting set and a learning set are constructed; the basic model carries out new knowledge learning on the learning set, and the training target of the model in the stage is that the accuracy of the learning set is close to 1, and meanwhile the accuracy of the reserved set is kept unchanged; after the basic model completes new knowledge learning, knowledge irrelevant to a new task is forgotten through a selective forgetting mechanism, and the training target of the model at the stage is that the accuracy rate of a forgetting set is close to 0, and meanwhile the accuracy rates of a reserved set and a learning set are kept unchanged. According to the method, continuous learning and machine forgetting are integrated together, so that the problem of contradiction between useless knowledge forgetting and new knowledge learning when a deep learning model processes knowledge updating of a pre-training model is solved. According to the knowledge exchange method based on first learning and then forgetting, the specified knowledge can be selectively forgotten while efficient learning of new knowledge is ensured, so that more refined knowledge regulation and control are realized, and the adaptability and the stability of the model are improved.

The invention relates to the technical field of unmanned aerial vehicle sensor anomaly detection methods, and particularly discloses an unmanned aerial vehicle flight control system data anomaly detection method based on a CNN-KFU multiple regression model. In unmanned aerial vehicle flight data, correlation of different degrees often exists among data, and firstly, a correlation analysis method is utilized to select a data set having relatively high correlation with target data; secondly, a multiple regression neural network model, namely CNN-KFU, for fine-grained spatial-temporal correlation analysis is designed based on 1D CNN and KFU, the CNN-KFU is used as a feature extractor of data, correlation of flight control data in time dimension and spatial dimension is fully learned, and the model is enabled to better understand data features. And finally, calculating the deviation between the regression residual error of the test data and a threshold value to realize abnormity judgment. The objective of the invention is to solve the problem that in the prior art, parameter selection is lack of effectiveness, and time-space relationship knowledge learning is insufficient, so that the anomaly detection capability is low in a complex flight scene.

An embodiment of the present application provides a method and system for accessing and processing integrated structured and unstructured data. By loading each template access session data into multiple deep learning units respectively, a confidence sequence of sensitive operation labels obtained for each template access session data under multiple deep learning units is generated. Based on the confidence sequence of sensitive operation labels of each template access session data and the sensitive operation annotation data corresponding to each template access session data, a sensitive recognition error value corresponding to each template access session data is determined. After selecting target template access session data from multiple template access session data, weight parameters of multiple deep learning units are optimized, and combined with multiple deep learning units to jointly perform sensitive recognition learning of access operations. Compared with using a single deep learning unit for knowledge learning, the deep learning effect can be significantly improved, and the accuracy of subsequent sensitive operation recognition can be improved.

The present disclosure relates to a drift detector to detect a drift in features of a diagnosis, the drift detector comprising a first collecting device configured to collect relevant explanations and raw data, a first database compiling the relevant explanations and raw data collected by the first collecting device, a second collecting device configured to collect relevant available expert knowledge from reliable sources, a second database compiling the relevant available expert knowledge from reliable sources collected by the second collecting device, a learned feature extractor configured to group initial data of the second database into a characteristic feature pattern for different diagnoses and to subsequently group the data from the first database into first feature patterns by diagnosis and from the second database into second feature patterns by diagnosis, a comparator configured to compare the empirical distributions of each of the grouped feature patterns in the databases with each other, and a processor configured to generate a report if results of the comparison do not comply with a predefined condition. Applications include medical applications such as recognizing new diseases.

This invention relates to the field of intelligent training technology, specifically to an internet-based intelligent training system for the prevention and control of respiratory infectious diseases. In this invention, a long short-term memory network is used to deeply model learners' learning progress, identify weaknesses in the learning process, and accurately push corresponding content, effectively providing personalized learning path recommendations. A graph neural network-based learning resource scheduling and content push method dynamically adjusts the frequency and order of learning content pushes to match learners' progress, ensuring that learning content is pushed at the appropriate time and with suitable difficulty. By constructing an interaction graph among learners and combining it with influence assessment, key prevention and control knowledge is disseminated preferentially through learners with high influence, optimizing the knowledge dissemination path. Combined with the application of the SIR (Self-Improving Influence) propagation model, the system predicts learners' progress in learning prevention and control knowledge, identifies learning bottlenecks, and significantly improves learning effectiveness.