22 results about "Layered learning" patented technology

The invention relates to the technical field of big data systems, and particularly discloses an AI big data driven hierarchical learning condition diagnosis and intervention system, which comprises a data acquisition module, a multi-dimensional learning condition fusion analysis module, an AI hierarchical diagnosis module, a dynamic intervention strategy generation module, an effect tracking module and a system linkage interface, through technical innovation, complete data, accurate diagnosis, precise intervention and smooth cooperation are realized in a K12 learning condition diagnosis and intervention scene, the problem that teachers are difficult to consider individual differences in large-scale teaching can be solved, the defect that high-quality teachers are insufficient in rural schools can be overcome, and a landing technical scheme is provided for K12 personalized teaching.

The invention discloses a robot motion control method and system based on cerebellum reinforcement learning, and the method comprises the steps: obtaining a current environment state vector, and inputting the current environment state vector to a main strategy channel and a cerebellum compensation channel in parallel; the main strategy channel outputs a basic action based on a long-term task target, and the cerebellum compensation channel outputs a compensation action responding to real-time dynamic through an efficient query mechanism; synthesizing the basic action vector and the compensation action vector into a synthesized action vector driving robot; feeding back latest data after the robot drives the motion action vector, and determining a sensory prediction error based on the latest data; and updating the original parameters of the cerebellum compensation channel based on the sensory prediction error, and optimizing the original strategy parameters of the main strategy channel based on the latest data. According to the invention, by constructing a parallel double-channel architecture, functional decoupling of advanced decision and rapid adaptation is realized, and unification of rapid adaptation and continuous optimization is realized through a double-loop hierarchical learning system.

The invention belongs to the technical field of power failure repair, and provides an intelligent generation method and system for a power failure repair scheme, and the method comprises the steps: constructing a multi-modal instruction template, obtaining the basic knowledge of the power industry, constructing a data set, training a general large language model through the data set, and employing a hierarchical learning rate strategy to achieve the intelligent generation of the power failure repair scheme. Obtaining a large language model after fine tuning of domain knowledge; obtaining a power failure first-aid repair case, constructing the power failure first-aid repair case as a sample according to a pre-constructed multi-modal instruction template, and optimizing the parameters of the large language model finely adjusted by the domain knowledge by using the sample by taking semantic loss, physical constraint loss and safety loss as a weighted mixed loss function; obtaining a large language model after multi-modal task fine tuning; and processing the target data by using the large language model subjected to multi-modal task fine tuning to obtain a final power failure repair scheme. According to the invention, the accuracy and intelligence of power fault first-aid repair scheme generation are improved.

The invention discloses an unmanned aerial vehicle data acquisition hierarchical learning method for a low-altitude wireless network. The method comprises the following steps: constructing a joint optimization problem by taking maximization of the total data volume collected from communication users during the flight of the unmanned aerial vehicle as a target; decomposing the joint optimization problem into a continuous unmanned aerial vehicle trajectory optimization sub-problem and a beam forming optimization sub-problem; for the unmanned aerial vehicle trajectory optimization sub-problem, a hybrid particle swarm optimization framework is adopted to search an optimal path; and for the beamforming optimization sub-problem, a deep learning model is used to obtain an unmanned aerial vehicle beamforming vector corresponding to each specified hovering position, and the input of the deep learning model is an estimated channel gain value between a communication user and an unmanned aerial vehicle. The method can effectively capture the coupling relation between the data collection demand and the motion of the unmanned aerial vehicle through the precise modeling data collection process and trajectory planning, and improves the overall collection efficiency and the resource utilization rate.

The invention discloses a task unloading and resource allocation method based on a mobile edge scene, which comprises the following steps of: a static decision-making stage: modeling a task unloading and resource allocation problem into a mixed integer nonlinear programming model, and aiming at minimizing the weighted sum of overall delay, overall energy consumption and individual delay; in the dynamic decision-making stage, a hierarchical learning framework is adopted, task unloading and resource allocation problems are decomposed, collaborative optimization is carried out through an upper-layer single agent and a lower-layer multi-agent, the upper-layer single agent is used for processing cloud data migration and resource allocation problems, and the lower-layer multi-agent is used for processing cloud data migration and resource allocation problems. And the lower-layer multi-agent is used for performing task unloading and base station selection on multiple users. The task unloading and resource allocation method based on hierarchical reinforcement learning is designed, the high-dimensional complex problem is more efficiently solved, and the task unloading and resource allocation optimization method is suitable for the task unloading and resource allocation optimization problem in a large-scale network.

The application discloses a kind of robot motion control method and system based on cerebellum reinforcement learning, it includes: obtaining current environment state vector, and parallel input to main strategy channel and cerebellum compensation channel;Main strategy channel outputs the basic action based on long-term task target, and cerebellum compensation channel then outputs compensation action by efficient query mechanism to cope with real-time dynamics;The basic action vector and compensation action vector are synthesized into synthesized action vector to drive robot;After robot drive movement action vector, feedback latest data, and determine sensory prediction error based on latest data;Based on sensory prediction error, the original parameters of cerebellum compensation channel are updated, and the original strategy parameters of main strategy channel are optimized based on latest data.The application realizes the functional decoupling of high-level decision and rapid adaptation by constructing parallel double-channel architecture, and the hierarchical learning system of double loop realizes the unity of rapid adaptation and continuous optimization.

This invention proposes a priority-based hierarchical learning method. For N tasks with different priorities, the evaluation function for task i is denoted as Q. i and maintain a prediction model π i For task i with priority, and all higher priority tasks j < i, prepare a predetermined threshold ε. ij and maintain a slack variable λ ij For any i > j, if Q j (π i )+ε ij <Q j (π j ), then represents π. i Performance on task j was better than π. j If the difference is too large, then increase λ. ij Conversely, λ decreases. ij But keep λ ij >0. With λ ij j < i is used as the weight to optimize π i Repeat the previous steps until convergence, and finally obtain π. N This is the desired model. In this invention, constraints are used to describe priorities, solving the problem of traditional multi-objective optimization methods lacking a priority order. Slack variables are introduced to automatically adjust the weights of each optimization objective. The dual variable is adaptively adjusted, resulting in zero duality with the primal problem, making it a convex optimization problem that can be solved quickly by existing solvers.

The invention relates to a partial discharge mode recognition method based on transfer learning. The method comprises the following steps: preprocessing original partial discharge pulse data into a PRPD spectrogram or a time-frequency image; the method comprises the following steps: constructing an EfficientNet-ECA model, embedding an ECA module into the model behind each convolution block of an EfficientNet-B0 network, dynamically carrying out channel dimension re-calibration on a feature map extracted by the EfficientNet-B0 network through the ECA module, and calculating an attention weight so as to re-scale the input feature map; migration training: adopting parameters of an ImageNet data set pre-training model to initialize network parameters of an OfficientNet-ECA model, and performing fine tuning through a hierarchical learning rate strategy; and inputting the PRPD spectrogram or the time-frequency image to be identified into the migrated and trained OfficientNet-ECA model, and outputting the partial discharge type to which the PRPD spectrogram or the time-frequency image belongs. Compared with the prior art, the method has the advantage of efficiently realizing accurate identification of the partial discharge mode.

The invention belongs to the field of behavior detection, and provides a skier falling detection method and system based on hierarchical learning, and the method comprises the steps: collecting multi-source motion data and personnel information of a skier, and carrying out the processing of the data, and obtaining time sequence feature data and personnel feature data; segmenting the time sequence characteristic data by using a window with a preset length, and marking according to a fall identifier to divide the time sequence characteristic data into fall data and non-fall data; screening non-tumble data by adopting a multi-level threshold method based on speed and attitude variation, and retaining data which are judged as high-risk behaviors; merging the falling data, the high-risk behavior data and data generated through data enhancement to form a model training data set; training the hierarchical learning network model by using the model training data set; and the trained hierarchical learning network model is utilized to identify real-time motion data, and skier falling detection is realized. The problems that in the prior art, generalization performance is poor, and individuation characteristics are not achieved are solved.

The application relates to a high-precision land carbon flux reconstruction method based on four-dimensional space-time deep learning, comprising the following steps: acquiring a multi-source heterogeneous data set and performing pretreatment; constructing a four-dimensional space-time deep learning model architecture connected in sequence from input to output, including a multi-scale space-time feature extraction module, a space-time feature fusion module, a prediction layer and an uncertainty quantification layer; taking the pretreated multi-source heterogeneous data set as input and taking a carbon flux reconstruction result as output; constructing a loss function, adopting a hierarchical learning rate strategy and introducing a cosine annealing scheduler, training the four-dimensional space-time deep learning model architecture based on the pretreated multi-source heterogeneous data set, and obtaining a four-dimensional space-time deep learning model; acquiring a carbon flux data reconstruction set and inputting the four-dimensional space-time deep learning model to generate a carbon flux reconstruction result. The method can effectively integrate multi-source heterogeneous data, capture the complex space-time dynamic characteristics of carbon flux, and realize accurate reconstruction of high-spatial and temporal resolution land carbon flux.

According to the microwave assembly heat source distribution optimization method based on the improved DNN algorithm, the improved DNN algorithm is applied to microwave assembly heat source distribution optimization, and a new optimization method is provided for microwave assembly heat source distribution optimization. According to the method, an improved DNN prediction model framework is constructed, residual connection, an attention mechanism and parallel CNN branches are fused in the framework, multiple groups of data are substituted into the framework, a hierarchical learning rate strategy is adopted for training, and model parameters are optimized by using a temperature constraint weighted loss function; according to the method, the structure of a model output layer is optimized, a dual-output form of'temperature mean value + prediction variance 'is adopted, and the uncertainty of prediction is quantified while the highest temperature prediction value is output; meanwhile, a composite loss function fusing a mean square error (MSE) and a temperature threshold penalty term is constructed, and for a prediction error exceeding a set safe temperature threshold, 2-3 times of weight is given to the loss function, so that the model preferentially guarantees the prediction precision of a high-temperature scene.

The application relates to the technical field of big data systems, and particularly discloses a layered learning situation diagnosis and intervention system based on AI big data driving, which comprises a data acquisition module, a multi-dimensional learning situation fusion analysis module, an AI layered diagnosis module, a dynamic intervention strategy generation module, an effect tracking module and a system linkage interface; through technical innovation, the application realizes full data, accurate diagnosis, precise intervention and collaborative order in the K12 learning situation diagnosis and intervention scene, can solve the pain point that teachers cannot give consideration to individual differences in large-class teaching, can make up for the short board of insufficient high-quality teachers in rural schools, and provides a feasible technical scheme for K12 personalized teaching.

This invention discloses a method and system for identifying malignant skin lesions based on an improved ConvNeXt network. Addressing the challenges of capturing subtle features in skin lesion images and the extreme imbalance between benign and malignant samples, this method enhances the model's feature perception capability of lesion regions from both channel and spatial dimensions by introducing an efficient channel attention module (ECA) and a spatial coordinate gating module (CoordGate) into the ConvNeXt network. The combination of Focal Loss and class weight design effectively mitigates the identification bias caused by the scarcity of malignant samples. Furthermore, transfer learning and hierarchical learning rate decay strategies are employed, loading pre-trained weights from a large number of samples to improve the model's convergence speed and generalization performance. This method significantly improves the identification accuracy and recall of malignant lesions while maintaining low computational complexity, demonstrating significant clinical application value.

The invention provides a learning behavior data analysis and report generation method for a multi-role education scene, and the method comprises the steps: obtaining original learning track data from three role terminals of teachers and parents of a management layer through a learning behavior data collection interface, and carrying out the classification marking of behavior records of different sources according to data source identifiers, segmenting the learning data according to three time granularities of day, week and month by adopting a time dimension slicing method to obtain a hierarchical learning data set with time labels; for a standardized multi-role report data set, a unique version identification code is distributed to each report through a version consistency management system, an automatic distribution mechanism is adopted to push the customized report to a personal workbench interface of a corresponding role user, and a personalized learning analysis report set supporting education decision is obtained.

The invention discloses a semantic hierarchical learning system based on a mesh DIKWP model, and the system builds a mesh semantic architecture with five layers of data, information, knowledge, intelligence and intention through a semantic graph layer dynamic perception module, a DIKWP-Learning engine (including a D / I / K / W / P-learning sub-module), a BUG recognition and backtracking module, an understanding closed-loop feedback system and a software and hardware cooperation platform. And bidirectional interaction and adaptive optimization of semantic information are realized. A BUG recognition and backtracking mechanism is innovatively introduced into the system, and learning bias errors are automatically corrected based on cognitive path tracking; and establishing an understanding closed-loop feedback system in combination with a relative consciousness theory, and dynamically adjusting a semantic layer weight to realize multi-subject semantic alignment. The system can be deployed in an edge computing device or a neural mimicry chip, has been applied in the fields of medical chronic disease intervention, judicial case retrieval, education personalized recommendation and the like, and effectively improves semantic comprehension accuracy and decision intelligence level.

The invention relates to the technical field of data processing, and discloses an intelligent data labeling method and system based on a multi-modal large model. The method comprises the following steps: receiving text, image and video data to generate a multi-modal embedding vector, carrying out annotation prediction and outputting a confidence score, extracting scene features from a low-confidence sample, clustering to generate a scene label, calculating gradient sensitivity of each network layer to construct a scene specificity parameter layer set, and carrying out parameter updating by adopting a hierarchical learning rate. And when the accuracy of the general scene is reduced and exceeds a threshold value, returning parameters and reducing the learning rate for re-updating. According to the method and the device, the accuracy of multi-modal data labeling in a complex scene and the adaptive capability of model parameter optimization are improved.

The application discloses a kind of based on layered learning index's multi-ndp equipment read-write acceleration method and system, method includes: in host end to key-value heat identification, in NDP equipment end executes key-value separation, respectively write hot value log area or cold value log area;Layered learning index is constructed and maintained, the value address of target key is positioned, and in multi-level cache collaborative management system is inquired;When the invalid data proportion of value log file reaches threshold, trigger garbage collection task.The application introduces learning index, accurately locates key-value address, converts multiple random reading of point query into small range sequential reading, significantly reduces read delay.Key-value separation and write heat identification are used to realize cold and hot data shunting writing, reduce write amplification.Multi-level collaborative cache mechanism based on migration benefit score is designed, dynamically optimizes the distribution of hot data between host and device, improves cache hit rate.Inefficient data is tracked and parallel recovery is triggered, improving storage space recovery efficiency, thereby systematically optimizing read-write performance.

This invention relates to a method, apparatus, and electronic device for high-precision entity recognition of railway work experience text, comprising: constructing a dedicated annotation system based on the characteristics of the railway human resources field, defining entity labels, including the start time of work experience, the end time of work experience, the unit and department, the position, the rank, and non-entities; acquiring a dataset of railway employees' work experience text, and manually annotating it according to the annotation system; training a named entity recognition model using the annotated dataset, the named entity recognition model being a model composed of a BERT model, a BiLSTM network, and a CRF network, fine-tuning the model parameters using a hierarchical learning rate, and introducing an early stopping mechanism during training; inputting the text to be recognized into the named entity recognition model, and outputting a named entity annotation sequence. This invention solves the problems of the lack of a targeted annotation system and suitable model in the railway field, as well as the insufficient recognition accuracy and incomplete entity coverage of existing methods in railway scenarios.

The invention relates to the technical field of computer vision, and discloses a fruit quality detection method and system based on MobileNetV2 and an automatic hierarchical learning rate. The core of the method is that when a MobileNetV2 model is adopted for fruit image quality detection, a hierarchical learning rate strategy is automatically implemented. According to the strategy, a model depth structure can be automatically analyzed, differentiated learning rates from low to high can be allocated to network layers with different depths without manual intervention, and a classifier layer automatically obtains a remarkably improved learning rate. Through the method, training is carried out in combination with gradient cutting and an early stop mechanism, and finally, the optimized model is deployed as an online detection system. According to the method, the training process is automatically optimized, the model training efficiency and the accuracy of fruit quality detection are remarkably improved, and meanwhile, the light weight and easy deployment of the model are ensured.

The invention discloses an enterprise embedded model training method and system based on an upstream and downstream relationship. The method comprises the steps of extracting and filtering out positive sample pairs meeting conditions from invoice data, generating negative samples, and segmenting the negative samples into a training set and a test set which are isolated from each other to form a sample set; the method comprises the steps that a sample set is set, after the sample set is processed, a vocabulary is formed, adjusted TinyBERT model architectures are trained, on the basis that the hierarchical learning rate is set and all-parameter training is adopted, a mean square error loss function is used for optimizing a double-tower model, and the double-tower model means that the two adjusted TinyBERT model architectures are used in parallel; and comprehensively evaluating the double-tower model according to the accuracy rate, the recall rate and the F1 value index, and storing the optimal double-tower model version based on the optimal F1 value. By implementing the method, training and reasoning efficiency can be remarkably improved, resource occupation is reduced, and the method is particularly suitable for efficient real-time processing and online low-delay reasoning requirements of large-scale enterprise data.