34 results about "Learning problem" patented technology

The invention discloses an engineering vehicle driver behavior analysis system based on federal learning. The system comprises a multi-source sensing layer which forms vehicle-road-person collaborative all-weather scenarized data input; the edge calculation layer is used for capturing a complex space relationship of a driving scene, embedding a lightweight cross-modal attention module, calculating an attention weight, and processing long-sequence driving operation data by adopting a Mamba state space model to capture a long-term dependency relationship of a driver state; the federated learning layer adopts a pFedMe algorithm to realize a driver customized model, during local training, the model learns local data distribution characteristics through a Moreau Envelopes regularization item, the cloud aggregates distribution of adaptive local data of each driver, models a federated learning process as a meta-learning problem, and performs training updating; a privacy and security layer; and differential privacy dynamic adjustment, Paillier improved homomorphic encryption and a Hyperledger Fabric block chain are adopted to prevent a log from being tampered.

The invention provides a data analysis method for personalized diagnosis and intervention of student learning questions, and the method comprises the following steps: S1, collecting an answer event sequence of a target student in a target question, the answer event sequence comprising an operation type and a corresponding timestamp; and S2, performing process segmentation on the answer event sequence based on a time interval of adjacent answer events, a submission event, a jump event and a rollback event. According to the method, process segmentation, step alignment and abnormal evidence extraction are carried out on the answering event sequence in the answering process of the student, and the learning abnormity of the student is specifically positioned to the corresponding standard question solving step and the trigger evidence thereof, so that a structured learning question positioning result is formed; therefore, a learning analysis result is converted from a single result index into traceable and explainable process diagnosis information, coarse-grained judgment only depending on an answer result or an abnormal score is avoided, and usability and pertinence of learning problem analysis are improved.

The invention discloses a continuous learning algorithm based on multi-task learning, which is used for solving the problems of real-time target detection and continuous learning of an unmanned system in a complex dynamic environment. The method comprises the steps that S1, an intelligent unmanned vehicle carries a high-precision sensor to collect multi-dimensional environment data, and importance samples are screened through a maximum gradient retrieval algorithm; s2, performing preliminary training on the pre-training model by using the screening data to enable the pre-training model to have basic target detection and recognition capability; s3, building a sea area image data enhancement continuous learning framework, and enhancing the image feature learning ability of the model under different weather conditions through an image compression reconstruction model, a cross attention module and an alternate training mode; s4, developing a stability and plasticity balancing strategy based on multi-task learning, relieving disastrous forgetting by solving a dual-objective optimization problem, and introducing a novel objective selection strategy to enhance the core set selection efficiency; s5, adding a regularization technology based on an influence function, and optimizing the performance of the model in the current environment; s6, in combination with a fine tuning technology, a general data pre-training model is firstly used, then the model is fine-tuned by using environment specific data, and the detection precision in a specific environment is improved; and S7, carrying out online learning and model evaluation optimization, continuously collecting new data to update the model, establishing a real-time evaluation and feedback mechanism, and continuously optimizing a target detection algorithm. According to the invention, powerful real-time target detection and continuous learning capabilities are provided for the unmanned system in a complex and changeable actual scene.

The application provides a password-based authentication key agreement method based on a lattice, which allows a user to use a password to agree with a server on a session key and authenticate the user identity. The method is constructed based on an ideal lattice, and its security is established on a ring-based learning problem with errors, so that the method can effectively resist quantum attacks. The user's credentials are saved in the form of password-encrypted in the server end, and only the user with the correct password can decrypt the legal credentials and establish the subsequent session key. When the session key is established, the user and the server perform an authenticated key exchange with a key hiding attribute to prevent the user's credential information from being leaked. Even if an enemy obtains the user's password-encrypted credential ciphertext and exhaustively searches the password space to decrypt the ciphertext to obtain a possible user credential set, the correct password corresponding to the credential cannot be identified from the set, so the application can resist offline dictionary attacks. In addition, two-way key confirmation can ensure the consistency of the session key and verify the user identity.

This invention discloses a data stream processing method and system based on deep reinforcement learning, relating to the field of industrial Internet of Things (IIoT) technology. It includes: S1: Constructing a topological graph neural network and determining the feature similarity weights of device topological features through a multi-head graph attention mechanism, dynamically aligning the device topological features of the source and target domains; S2: Setting up a main channel through a deep reinforcement learning model and an auxiliary channel through a temporal contrastive meta-learning model to obtain device health assessment and fault mode features; S3: Monitoring the data stream based on the health assessment and fault mode features, and reorganizing the network structure through a neural architecture search model. This invention solves the transfer learning problem caused by device heterogeneity in the IIoT, dynamically balancing physical distance, process coupling, and feature similarity, improving the initial accuracy of the source domain model in the target domain, and reducing the amount of labeled data required in the target domain.

The invention discloses an open intelligent algorithm adaptive combination method and system based on meta learning and a dynamic pipeline. According to the method, by introducing an algorithm directed acyclic graph construction mechanism of a type system and an adapter, dynamic integration of heterogeneous components in an open algorithm library is realized, and the limitation that a traditional AutoML system algorithm combination space is closed and depends on a preset process is overcome. Environment perception and a performance monitoring closed loop during operation are integrated into a decision process, and a dynamic adjustment mechanism based on deviation triggering is designed, so that the system has the capability of coping with concept drift and resource fluctuation after deployment. By constructing and continuously updating the knowledge base recording the task-environment-performance mapping relation, the system has the ability of making decisions based on historical experience. According to the meta-learning mechanism, when the system processes similar tasks, repeated search overhead can be reduced, continuous evolution of performance is achieved, and a feasible technical path is provided for solving the continuous learning problem of a universal intelligent system.

We provide the system. [Solution] A means for acquiring the user's learning history information and selecting learning problems optimized for the user, A means for dynamically generating dialogue content based on the selected learning questions, A means for receiving user voice input, converting it into text data, and analyzing the text data, A means of providing feedback to users based on the analysis results, A means for detecting a user's mental stress and providing dialogue content for mental care as needed, A means of recording the user's learning results and updating the learning profile for the next session, A system that includes this.

The invention relates to the technical field of e-government affairs, in particular to an intelligent grassroots party building system which comprises a table body, and a convenient assembly is arranged at the top of the table body. According to the intelligent grassroots party building system, by installing a convenient assembly, when students submit own life or learning problems online, a stool at the bottom of a table body can be pulled out, then an adjusting block is pulled, so that a limiting column can be moved out of an adjusting hole, then the stool can be stretched, and the length of a telescopic rod can be adjusted; after the stool is adjusted to a proper height, the limiting columns can be inserted into the specified adjusting holes, and the height of the stool can be adjusted to be the height suitable for students, so that the sitting posture is more comfortable through the more proper height when an opinion is submitted or communication is carried out, the students can concentrate on more attention, and meanwhile, after the communication is finished, the students can conveniently and quickly sit on the stool. And the stool can be placed on the fixed sliding rod in the storage groove, so that the stool can be stored, and the occupied space is reduced.

The invention provides an intelligent tutoring system driven by an LLM-based collaborative intelligent agent. The intelligent agent innovatively utilizes a teaching strategy to perceive a discussion stage, discover learning problems, determine an intervention opportunity and generate instructive feedback. The invention provides a novel prompting strategy based on exploration of a community theory, so as to cultivate an intelligent agent to understand the overall discussion situation and timely discover problems encountered by each student. In order to discover and learn unknown influence of feedback of an AI tutor on subsequent collaborative learning activities of students, the invention provides a multi-agent framework, and collaborative learning behaviors between a tutor agent and a plurality of student agents are simulated by automatically executing collaborative learning tasks. Meanwhile, synthetic learning data is generated to further finely tune the LLM, and the influence of an AI tutor on the cooperation behavior of the students is learned, so that instructive feedback is generated in time under the condition that the cooperation of the students is not interfered. Finally, instructive feedback can be generated at a proper time to support collaborative learning of students.

The application discloses a stamping process intelligent decision-making method of man-machine cooperation, constructs a stamping forming prediction model based on a physical information neural network, embeds Hill48 yield criterion, Swift hardening model and friction law into a loss function in the form of a regular term, trains process sensitivity gradient by using small sample data and outputs the process sensitivity gradient, calculates information gain by using an expectation improvement or a confidence upper bound acquisition function and recommends a die trial point based on model cognitive uncertainty, triggers manual confirmation when uncertainty is higher than a threshold value, executes process parameters, collects a stamping force-displacement curve in an initial die trial stage, inverses sheet metal parameters by using an extended Kalman filter to correct model input, records results and feeds back and updates the model, solves a small sample learning problem by embedding physical constraints, reduces die trial times by active learning, realizes material self-adaptation by online inversion, and forms a closed-loop optimization of prediction-die trial-inversion-updating.

The application discloses an emergency protection site multi-target location optimization method based on integrated reinforcement learning, belongs to the field of site location optimization, and comprises the following steps: S1, acquiring candidate points and demand point data in a region, and constructing a multi-target reinforcement learning problem; S2, performing Markov modeling based on a single-agent decision deep reinforcement learning problem; S3, constructing an integrated reinforcement learning framework; S4, training the constructed integrated reinforcement learning network; and S5, obtaining a location result by using the integrated reinforcement learning framework and performing visual presentation; the application introduces an integrated learning strategy and an adaptive optimization technology, combines spatial information technology of a remote sensing and a geographic information system core, realizes multi-target location optimization of urban emergency protection sites, and enhances robustness and practicability of an aggregated reinforcement model of an Actor-Critic network.

The invention provides an intelligent information processing method for family-school cooperative cultivation. The method is applied to the technical field of data processing. The method comprises the steps of preprocessing multi-modal data; obtaining a student learning semantic vector according to the extracted single-mode characteristics of the preprocessed multi-mode data; a knowledge graph is constructed based on the student learning semantic vector, and a personalized ability vector is generated in combination with student historical learning data; calculating the matching similarity between the personalized capability vector and the educational resource vector, screening a plurality of educational resources with the highest matching degree, and pushing the educational resources to a corresponding terminal; predicting a learning problem of the student in a future preset period through an LSTM time sequence model, and generating an intervention strategy in combination with the knowledge graph; receiving demand information of home and school users, determining demand intentions and key entities, and generating targeted response information based on the personalized capability vector; and performing access control on the processed multi-modal data and the student learning semantic vector. In this way, the information processing efficiency can be improved.

According to the correction scheme recommendation method based on small sample learning, the meta-learning technology is introduced to solve the small sample learning problem of correction measure recommendation of prisoners serving prisoners, the incidence relation between prisoners and criminals can be more fully modeled, and correction measure recommendation strategies of small sample groups are more effectively mined; a meta-learning method is used, hierarchical information is used, and the capacity of obtaining parameters matched with tasks through internal circulation gradient descent of initial parameters is enhanced; by considering priori knowledge of prisons and criminals, modeling is carried out on the incidence relation on the initial parameter level, and the capacity and rationality of processing task space heterogeneity are enhanced; the globally shared internal loop learning rate is scaled, and a more reasonable internal loop learning rate is obtained according to the relative semantic information, so that the internal loop gradient descent can better obtain parameters adapted to the task; and more appropriate global parameters can be obtained according to the learning ability of each task support set when the global parameters are updated in the outer circulation.

This application relates to the field of hybrid solid-state drive (SSD) technology, and discloses a hybrid SSD space management method and apparatus based on multi-agent reinforcement learning. This method models the coupled garbage collection and flash mode switching operations in a hybrid SSD as a reinforcement learning problem decided by independent agents. A multi-agent reinforcement learning scheduler is added. By monitoring the shared system status in real time, such as the SLC space ratio and write frequency, two agents independently select the optimal action from an action space containing cross-region garbage collection and mode degradation options, and combine them into a joint command for execution. Finally, the scheduling strategy is dynamically adjusted based on a differentiated reward function that balances performance metrics and internal write overhead. This invention solves the problem that the two background operations, garbage collection and mode switching, are executed independently and compete for shared resources, easily interfering with each other, leading to decreased front-end I / O performance and a sharp increase in write latency.

The application discloses a cementing material production energy consumption evaluation method based on Internet of Things collection, comprising a meta-learning and transfer learning fusion framework, and is characterized in that the method further comprises the following steps: S1: the meta-learning and transfer learning fusion framework first divides source domain cementing material production data into batches and classification tasks as basic units of training cementing material production data, and the cementing material production data divided in the task unit mode is used as a basic input unit of meta-learning. Through the multi-task learning paradigm of meta-learning, the division of the support set and the query set is utilized to realize rapid adaptive learning of the model under the condition of a small amount of cementing material production data samples, solve the small sample learning problem, in addition, the knowledge transfer from the source domain cementing material production data to the target domain data is realized through the transfer learning, the cross-domain learning problem is solved, and the adaptability of the model to different production environments is enhanced.

We provide the system. [Solution] A means for users to take pictures of and upload learning problems they answered incorrectly, An image analysis means that analyzes uploaded images and extracts text data, An analytical method for identifying a user's weak areas from extracted text data, A problem generation method that generates and provides similar problems based on the user's areas of weakness, A means of providing explanations that prepare and present to users explanations for similar problems that have been generated, A feedback system for recording learning progress and creating new learning plans, A system that includes this.

The utility model discloses an intelligent learning monitoring device, which comprises a host and a camera, the camera can support horizontal rotation and vertical rotation, the host is provided with a coding compression module, a network transmission module, a coding playing module, a storage module and a display screen, a base is arranged below the host, the base is connected with the host through a connecting mechanism, and the host is connected with the network transmission module. A charging panel is arranged on the base, a receiving module is arranged in the host, the camera can be checked by a monitoring person through the arrangement of a network transmission module, so that the learning condition of a learner is observed, and the learner is helped to solve the learning problem through video connection with the learner through a display screen; through the arrangement of the base, the contact area between the device and a placement plane can be increased, the stability is improved, meanwhile, through the arrangement of the charging panel, the device can be charged in real time during use, and therefore the situation that the device is charged by changing the position of the device is avoided.

This invention discloses a federated learning method and system based on unified representation and classifier correction, designed to address the federated learning problem under long-tailed data distributions. First, this invention obtains a globally unified prototype by aggregating local prototypes (i.e., the average features of categories extracted by the global model). These prototypes are then used to adjust the feature space, bringing features within the same category closer to the corresponding globally unified prototype while pushing away features from other categories. Furthermore, this invention reduces classifier bias through prototype mixing using the global prototype. It generates a balanced virtual feature set by fusing the globally unified prototype and local features. The classifier is then retrained on this feature set to correct the decision boundary and mitigate bias. This invention can effectively improve the classification performance of models under long-tailed data while reducing the bias in the feature space and the classifier. The superior performance of the method has been verified on multiple datasets.

The application discloses the technical field of coding strategy and relates to a multi-instance multi-label learning method based on a combined error correction coding strategy. The multi-instance multi-label learning method based on the combined error correction coding strategy comprises the following steps: a feature embedding representation method of multi-instance bag level data structure information is designed based on a Fisher kernel, and an original multi-instance multi-label learning problem is converted into a multi-label learning problem; under the guidance of an error correction output coding idea, a coding strategy of randomly connecting multi-label sub-problems is proposed; a support vector machine model is constructed and trained, and hard labels representing the specific category performance of a predicted sample under each label are obtained based on a T criterion. The multi-instance multi-label learning method based on the combined error correction coding strategy effectively alleviates the class imbalance problem. In addition, the model divides different training data blocks based on coding random division to learn base classifiers, so that the similarity between the base classifiers is weakened, and the classification coding error correction capability is stronger.

This invention relates to a method and apparatus for training a neural network model on noisy multi-label data. The method includes the following steps: selecting a clean set of samples for each category as a meta-dataset using a sample selection algorithm, and estimating the category-dependent label noise transition matrix; initializing some parameters in the instance feature-dependent label noise transition matrix network using the category-dependent label noise transition matrix; transforming the learning problem into a two-layer optimization problem based on statistically consistent label noise learning loss, and simultaneously learning the instance feature-dependent label noise transition matrix network parameters, data imbalance parameters, and multi-label classification neural network parameters using a meta-learning algorithm. This invention innovatively utilizes a meta-learning algorithm in a data-driven manner to unify the learning of instance feature-dependent label noise transition matrix network parameters, data imbalance parameters, and multi-label classification neural network parameters within a single framework.

The application provides a multi-UGV path planning method and device based on multi-agent reinforcement learning. The method comprises steps 1 to 8. The application operates on the basis of existing resources, improves mutual cooperation and coordination between each agent in the multi-agent system, uses a distributed search structure, automatically decomposes a complex learning problem into a local sub-problem that is easier to learn, improves the intelligent degree, expands the application field, learns a decentralized strategy in a centralized setting, greatly reduces the calculation amount, and has high practicability.

The invention provides a security reinforcement learning control method and system for an average residence time switching system based on multiple Lyapunov stability constraints, and belongs to the field of intelligent control and reinforcement learning. The objective of the invention is to solve the problem of how to introduce a multi-Lyapunov stability criterion suitable for a switching system in reinforcement learning and realize stable and reliable strategy learning on the premise of satisfying an average residence time constraint. According to the method, a mode-dependent Lyapunov function is constructed for each subsystem mode, a multi-Lyapunov stability criterion meeting an average residence time condition is introduced, and a multi-Lyapunov descent constraint is explicitly applied in a reinforcement learning strategy updating process; therefore, the mean square index stability of the switching system in the learning stage and the execution stage is realized. The method does not need to depend on an accurate system dynamics model, can safely learn a mode-dependent control strategy only based on the state and control data collected in the system operation process, and has high robustness and engineering applicability.

The application discloses an online learning input recognition method based on multi-vision clue fusion. Firstly, the application faces the demand of large-scale online learning input perception, starts from the multi-vision clue angle, excavates the associated vision clues of the online learning input, and constructs a multi-dimensional fine-grained representation model of the online input. Secondly, the feature learning problem of the time sequence is converted into a graph-based feature learning problem, a graph network model based on mutual information regularization is proposed, meanwhile, training support is provided for the machine learning method used by the application, and a learning input perception database based on multi-vision clues is constructed. Finally, a fine-grained learning input recognition method fusing multi-vision clues is constructed, and on this basis, a coarse-grained learning input recognition method based on the input graph is designed to integrate the fine-grained variable-length learning input sequence, so that multi-granularity online learning input recognition is finally realized, and the multi-level and multi-stage learning input perception demand in actual application is met.

The invention discloses a training method and application of a humanoid robot motion control model, and belongs to the technical field of robot motion control. And reward learning is modeled into a meta-learning problem, and joint optimization of a control strategy and a reward function is realized through alternate iteration of an upper-layer optimization process and a lower-layer optimization process. Wherein in the lower-layer training, a motion control strategy of the robot is learned through a reinforcement learning algorithm, and a strategy network outputs a control action based on part of observation information which can be acquired by the robot and is optimized through a predicted reward value provided by an upper-layer reward model; and meanwhile, the upper layer reward model evaluates the strategy behavior by using the complete state information of the robot, so that the intelligent agent is guided to form a coordinated, stable and physically consistent motion mode on the global level. Through the above process, the reward model can adaptively evolve along with the strategy learning process, so that the humanoid robot is guided to efficiently and stably learn the complex motion control strategy without manual intervention.

The application discloses a deep multi-label classification network robust training method based on an information entropy maximization regularization mechanism, aims to solve the robust learning problem of a deep multi-label classification network under the condition that each sample has only a single positive label and the rest of the labels are all missing, solves the SPML problem by using a binary cross-entropy loss function regularization mechanism based on information entropy maximization, aims to maximize the information entropy of the prediction probability of the unknown label of the deep multi-label classification network, so that the model can be immune to false negative label noise, learns by using the supervision information provided by the real positive label, and makes a prediction with strong distinguishability for the unknown label, the method is easy to implement, does not introduce additional learnable parameters, can be combined with any deep multi-label classification network, and is suitable for processing super large-scale multi-label image data sets.

The application belongs to the field of graph representation learning, and proposes a multi-agent hypergraph modeling and representation method. First, the graph representation learning process is defined as a multi-agent optimization process; second, the message passing mechanism in the existing hypergraph representation learning method is taken as an interaction mode of an agent; then, a force-based agent interaction mode is designed; finally, the two agent interaction modes are optimized together on the downstream task, and the optimized graph model is used to output the prediction result. The application formalizes the graph representation learning problem from the perspective of multi-agent, considers the high-order complex relationship of nodes in the graph, and combines message passing and force-based interaction, which improves the performance of the existing hypergraph method and is suitable for graph learning tasks in complex scenarios.

The invention belongs to the technical field of artificial intelligence and robots, and discloses a structured learning method of an artificial intelligence robot, which comprises the following steps of: 1, multi-semantic decoupling and independent embedding: carrying out structured isolation and decoupling on semantic information related to robot tasks, and creating an independent embedding channel and a learning sub-module for each type of semantics; 2, a content-aware mixed attention mechanism: aiming at different semantic module characteristics, designing the content-aware mixed attention mechanism, and dynamically adjusting an attention interaction strategy between different semantics; through semantic decoupling, complex tasks are decomposed, the learning problem is simplified, rapid convergence of the model is facilitated, and the requirement for the data size can be possibly reduced; the content-aware attention mechanism effectively reduces irrelevant information interference, so that the model can utilize key information more efficiently, and higher precision and robustness are obtained on each sub-task.

The invention discloses an education and teaching auxiliary method and system based on big data, and belongs to the technical field of data analysis, and the method comprises the following steps: S1, extracting education and teaching data from a big data source in real time; s2, constructing a knowledge graph based on the education and teaching data, and presetting an adjustment unit in the knowledge graph; s3, collecting learning behavior data of students in real time, generating weak nodes through an adjusting unit, and calculating influence factors according to the weak nodes; s4, adjusting association weights of nodes in the knowledge graph based on the influence factors, and generating a reconstructed knowledge graph; and S5, based on the reconstructed knowledge graph, generating a learning path and tutoring content. The system comprises a data acquisition module, a knowledge graph engine, a student model analysis module, a knowledge graph reconstruction module and a personalized tutoring module. According to the method, the learning behavior data is collected in real time and is deeply associated with the knowledge graph, so that the accuracy and pertinence of learning problem diagnosis are remarkably improved.

This application relates to the field of smart teaching technology, and in particular to a method, device, medium, and product for tracking student cognitive paths and recommending teaching. The method includes: acquiring the answer data of a target student in a target assessment stage, where the target student is any student in the target class; accessing the teaching knowledge graph of the target assessment stage, where nodes represent knowledge points and edges represent dependencies between knowledge points; mapping the answer data to the teaching knowledge graph, identifying implicit pre-breakpoints, and constructing a dynamic knowledge graph of the target student based on these implicit pre-breakpoints; aggregating the dynamic knowledge graphs of all students in the target class to generate a list of common weak knowledge points; and pushing the list of common weak knowledge points to the teacher's end to guide the teacher in conducting focused teaching. This application can achieve accurate tracking of the cognitive path of an individual student and can also extract common learning problems at the group level, improving the relevance of teaching.