52 results about "Random projection" patented technology

The invention discloses a magnetic method data physical property inversion method and system based on a deep learning terrain disturbance layer, and relates to the technical field of geophysical exploration, and the method comprises the steps: building a magnetic anomaly data set, generating an underground three-dimensional magnetic anomaly model through a mixed hexahedron model and a terrain disturbance function, and obtaining corresponding magnetic anomaly data through forward modeling, each group of magnetic anomaly data corresponds to different terrain parameter combinations; constructing a deep learning inversion network model based on UNet; and training the deep learning inversion network model based on the magnetic anomaly data, taking the trained training set parameters as the weight of the deep learning inversion network, verifying the effect by using the verification set after the training is finished, and predicting the effect according to the test set. According to the method, a traditional same-dimensional random projection inversion method is combined with a deep learning network structure, and dynamic transformation of terrain parameters is realized by using a terrain disturbance layer, so that the network keeps high inversion precision and stability under different terrain fluctuation conditions.

The steam generator state real-time monitoring method based on artificial intelligence belongs to the field of artificial intelligence and comprises the following steps: S1, data acquisition and labeling; S2, sample generation is performed by using a quantum generative adversarial network based on random projection embedding to realize data expansion; S3, the expanded data is input into a feature extraction model to perform training of the feature extraction model, and a five-layer fully connected neural network is used for feature extraction; S4, the feature-extracted data is input into a feature dimension reduction model to perform training of the feature dimension reduction model, and a self-encoding neural network algorithm based on local preserving projection is used to realize feature dimension reduction; S5, the dimension-reduced data is input into a classifier to perform training of the classifier model; and S6, steam generator state recognition and monitoring are performed.The steam generator state real-time monitoring method based on artificial intelligence can solve the problems of insufficient sample quantity and lack of data diversity and enhances the robustness of the model when the model has noise or fuzzy classification boundary data.

The invention discloses a sales supply chain management method and system based on Internet data, and relates to the technical field of supply chain optimization, and the method comprises the steps: carrying out the decomposition through discrete wavelet transform, calculating an approximate kernel distance through Gaussian random projection, generating low-dimensional embedding through low-rank decomposition and ADMM iterative optimization, and initializing the prediction demand of Poisson distribution. The method comprises the following steps: quantifying through a Shapley value, initializing a particle swarm, obtaining an individual and global optimal position, carrying out local guide search, outputting an optimal scheme through domain operation for updating, and generating an optimal logistics path and a scheduling scheme through SA disturbance and DE-GA variation. According to the method, the prediction precision is improved by introducing Poisson distribution and Shapley value quantification prediction requirements, an efficient logistics path and a scheduling scheme are generated through multi-target particle swarm optimization and a hybrid variation strategy, and the adaptability and the operation efficiency of a sales supply chain in a dynamic market environment are improved.

The invention relates to a lightweight airborne visual tracking method and device based on random projection and a medium, belongs to the technical field of airborne computer vision and target tracking, and can reduce the calculation and storage cost of airborne platform feature processing, guarantee the target tracking precision and adapt to resource limited scenes. According to the invention, lightweight airborne visual tracking is realized based on locality sensitive hash (LSH) random projection, the method is suitable for resource-limited scenes of airborne platforms such as unmanned aerial vehicles and helicopters, and typical challenges such as target scale dramatic change, complex background interference and high-altitude high-angle shooting visual angle offset under airborne vision can be efficiently processed. The core requirements of an airborne system on low computing power occupation, low storage consumption and high real-time performance of a tracking algorithm are met; the LSH random projection technology is organically combined with an existing lightweight tracking architecture, on the premise that tracking precision is not remarkably lost, the calculation and storage cost of feature processing is further reduced, and the method becomes a constraint of breaking through airborne visual tracking resources-performance.

The invention relates to the technical field of information security, in particular to a federal data privacy protection method based on self-adaptive sparse homomorphic encryption, which comprises the following steps: firstly, carrying out local model training and calculating weight parameter difference at a client, and carrying out parameter block screening through a dynamic sparse proportion to generate a mask vector; only encrypting the reserved parameter block by adopting a CKKS homomorphic encryption algorithm, and generating a low-dimensional weight for contribution degree evaluation through random projection; and the server side calculates a contribution weight based on the cosine similarity of the projection weight and the historical accuracy, and completes weighted aggregation in a ciphertext state. According to the method, through a sparse-encryption-projection triple protection mechanism, the communication overhead is remarkably reduced while the model precision is ensured, the gradient leakage risk is effectively prevented, and the method is suitable for distributed modeling scenes of sensitive data of medical treatment, finance and the like.

The application discloses a medical image class incremental learning method and system based on feature principal direction guided projection, and belongs to the technical field of medical image analysis; the method acquires medical image features through a pre-trained feature extractor; random projection is performed on the features to obtain first-view features; a covariance matrix is calculated based on a feature cache area and is decomposed to extract the first K principal directions, construct a guided projection matrix, and obtain second-view features; the double-view features are spliced to obtain final representation; the classifier weight is directly calculated through a ridge regression closed-form solution without backward propagation training; the application adaptively captures the core structure of the feature space through data-driven principal direction projection, retains the detailed information in combination with random projection, effectively improves the feature separability of fine-grained classes, adopts an analytical classifier updating mechanism, realizes stable class incremental learning without storing historical data, improves the model updating efficiency, and is suitable for efficient continuous learning tasks in various medical image scenes.

Embodiments of the present application provide an alarm storm suppression method and device, computer storage medium and product. The alarm storm suppression method comprises: obtaining alarm data corresponding to an alarm storm; repeatedly randomly sampling the alarm data and constructing an isolation tree based on any preset segmentation method for the alarm data obtained in each sampling, wherein the preset segmentation method comprises a first preset segmentation method and a second preset segmentation method, the first preset segmentation method is to perform splitting based on a numerical feature and a preset segmentation threshold, and the second preset segmentation method is to perform splitting based on a random projection corresponding to a semantic feature; adjusting a path length of the alarm data in each isolation tree based on an alarm device topology weight corresponding to each alarm data, determining an anomaly score of the alarm data based on the adjusted path length; and determining alarm data with an anomaly score exceeding a preset anomaly score threshold as core alarm data. According to the method provided by the embodiments of the present application, the efficiency and accuracy of alarm positioning are improved.

The application discloses a magnetic data physical property inversion method and system based on a deep learning terrain disturbance layer, relates to the technical field of geophysical exploration, and comprises the following steps: establishing a magnetic anomaly data set, generating an underground three-dimensional magnetic anomaly model by mixing a hexahedron model and a terrain disturbance function, and obtaining corresponding magnetic anomaly data by forward calculation, wherein each group of magnetic anomaly data corresponds to different terrain parameter combinations; a deep learning inversion network model based on UNet is constructed; the deep learning inversion network model is trained based on the magnetic anomaly data, the trained training set parameters are used as the deep learning inversion network weight, the effect is verified by using a verification set after the training is completed, and the prediction effect is predicted according to a test set. The application combines a traditional same-dimension random projection inversion method and a deep learning network structure, realizes dynamic transformation of terrain parameters by using a terrain disturbance layer, and enables the network to maintain high inversion precision and stability under different terrain undulation conditions.

The invention discloses a distributed sparse model fingerprint method based on dual-key driving, and the method comprises the following steps: inputting an original model, an owner secret key and an identity label into a computer system; generating a seed based on an owner secret key, calculating a global index set in one or more stable target convolution layers pre-selected in the model, and selecting a weight subset as a fingerprint carrier; constructing a random projection matrix based on the identity label; the weight vector and the random projection matrix generate an original model fingerprint; verifying the suspicious model, repeating the steps to obtain a fingerprint to be detected, and calculating the cosine similarity between the original model fingerprint and the fingerprint to be detected; and judging whether an illegal derivative relationship is formed based on a calibrated threshold value, and outputting an ownership judgment result. According to the method, identity and pseudo-random processes are bound through double keys, and distributed sparse sampling and normalized random projection are combined, so that the uniqueness and robustness of fingerprints are improved, various attack scenes can be effectively resisted, and the ownership of the model is accurately judged.

The application discloses a kind of based on big language model's evolvable software vulnerability detection method, comprising: by regular pattern matching identification Source sentence, based on call graph traversal and data dependence analysis execution function level inter-process slice, build cross-function code context, input the big language model of parameter efficient fine-tuning, output the vulnerability propagation path from Source to Sink;When new vulnerability type needs to be extended, the parameter variation characteristics of old data are extracted by multi-step fine-tuning, and the representative core set is selected by random projection dimension reduction and hybrid distance hierarchical clustering, and the training is played back by mixing new data to alleviate catastrophic forgetting;In the actual use process of tool, the false alarm and the false alarm confirmed by user are collected as feedback signal, the core set is clustered and layered filtered and refined based on perplexity and error prediction analysis, the harmful old knowledge that leads to false alarm and false alarm is removed, and verified new mode is supplemented at the same time, to realize the closed-loop evolution of self-improvement.

This invention relates to the field of wind and solar power engineering construction and ecological restoration technology, and discloses a method and system for gravel curtain layer restoration based on multimodal data fusion. The method includes the following steps: acquiring multimodal data including natural environment, engineering disturbance, and remote sensing data, and independently encoding them into feature vectors of a unified dimension; constructing a bidirectional cross-modal attention mechanism to fuse natural environment and engineering disturbance features; constructing a sequence by combining the fused features and remote sensing features, and inputting it into a multi-head self-attention network for joint modeling; extracting high-order features through random projection expansion and third-order interactive operations, and inputting the extracted features into a residual feedforward discrimination network to output the damage level; and matching and outputting restoration schemes from a pre-built scheme library based on environmental and engineering constraints. This invention achieves structured fusion of multi-source information and intelligent decision-making, improving the accuracy and engineering adaptability of gravel curtain layer restoration.

The application discloses a kind of based on random projection's physical property parameter inversion method of potential field data, comprising the following steps: S1: measured potential field data is obtained, according to survey area and depth range is profiled in underground space, sensitivity matrix in inversion is calculated based on potential field data forward theory, and then the forward calculation relationship of full space is established;S2: random projection matrix is designed, and sensitivity matrix is projected to multiple low-dimensional subspace, and the subspace forward calculation relationship is established;S3: based on the regularization equation of subspace forward calculation relationship, and the physical property parameter in subspace is solved using conjugate gradient algorithm;S4: the final physical property parameter inversion result is obtained by the weighted average calculation of multiple physical property parameters in subspace.The physical property parameter inversion method of potential field data based on random projection has higher depth resolution and inversion reliability, and improves the practicability of physical property inversion method in actual data processing.

The invention relates to the technical field of natural language processing, in particular to a model prefix parameter and hyper-parameter joint optimization method and system. Comprising the following steps: S1, acquiring a high-dimensional prefix parameter and hyper-parameter joint optimization instruction; s2, generating a random projection matrix, and mapping the high-dimensional prefix parameters into low-dimensional prefix parameters through the projection matrix; s3, searching an optimal low-dimensional prefix parameter in a low-dimensional space through a covariance matrix adaptive evolution strategy, reconstructing the low-dimensional prefix parameter into a dynamic high-dimensional MLP parameter, generating a key-value prefix required by each layer of the model, and embedding and injecting the key-value prefix into the large model; s4, training hyper-parameters are dynamically adjusted based on a particle swarm algorithm and a cosine annealing strategy, and the hyper-parameters are transited step by step in the training process; s5, training the model based on the optimal prefix parameter and the hyper-parameter, and outputting the trained model; and S6, applying the trained model to the medical question and answer task. According to the method, the precision and stability of prefix tuning in the medical question and answer task can be improved.

The invention discloses a remote abnormity monitoring electric energy meter with an edge diagnosis function and a monitoring method. The method comprises the following steps: acquiring preprocessed voltage and current synchronous sampling data; generating a current electric energy parameter feature vector through a sliding window and a multi-time scale strategy; outputting a random projection forest leaf node set; calling an extreme learning machine local classifier bound with each leaf node one by one, and performing local reasoning by using the current electric energy parameter feature vector to obtain a corresponding local discrimination score set; and outputting an abnormal power fusion confidence coefficient, and comparing the abnormal power fusion confidence coefficient with an adaptive threshold to generate an abnormal power identification result. According to the invention, the expression and discrimination capability of the inner side of the electric energy meter on diversified abnormal power states is greatly enhanced.

The invention discloses an adaptive pulse parallel method and system for efficient training of a pulse neural network. The method comprises the following steps: initializing a shared weight and a fixed random projection matrix; fragmenting the data and locally accumulating the gradient; calculating a gradient consistency direction after synchronization; generating an adaptive aggregation weight based on the cosine similarity and a Softmax function; weighting the aggregation gradient and synchronously updating the global weight; and dynamically unloading the operator to the CPU or the GPU according to the historical performance during forward propagation. The invention aims to solve the problems of insufficient parallelization capability, high communication overhead and low heterogeneous resource utilization rate in neural network training in the prior art, and realizes efficient parallel training of the pulse neural network without back propagation by combining the pure feed-forward characteristic of the DRTP algorithm, the adaptive gradient aggregation strategy and the adaptive element-by-element unloading mechanism.

Systems and methods for generating single-node representations in graphs composed of linked nodes. The present technology enables dynamic generation of individual node embeddings in sub-linear time (less than O(n), where n is the number of nodes in the graph G), using only the PPR vectors of the individual nodes and a random projection to reduce the dimensionality of the PPR vectors of that node. In one example, the present technology includes a computer-implemented method comprising: obtaining, from a database, a graph having a plurality of nodes; generating, for a given node of the plurality of nodes, a personal PageRank vector; and producing, for the given node, an embedding vector by randomly projecting the personal PageRank vector, wherein the embedding vector has a lower dimensionality than the personal PageRank vector.

The application discloses a fast calculation method and device for scattered electromagnetic field based on random matrix approximation, and belongs to the field of fast modeling of scattered field. The method comprises the following steps: precalculating an incident field vector and a reciprocal incident field vector; generating a Gaussian random matrix matched with the column number of a to-be-determined scattered response matrix; performing linear superposition on the incident field vector to obtain a synthetic incident field vector, and constructing a random projection matrix according to the scattered field vector of an equivalent current vector at a receiving point; performing linear superposition on the reciprocal incident field vector to obtain a reciprocal synthetic incident field vector, and constructing a projection coefficient matrix according to the response of a reciprocal equivalent current vector at a transmitting source; and calculating a complete scattered response matrix based on the product of an orthogonal basis matrix and the projection coefficient matrix. In the application, a large-scale geophysical electromagnetic complex scene can be adapted, higher calculation precision and stable convergence are achieved, and the reconstruction of the complete scattered response matrix can be completed at a lower relative error level.

The invention provides a reinforcement learning data selection method and device based on off-line strategy influence estimation. According to the method, firstly, offline approximation of a target strategy gradient is realized through importance sampling and KL divergence constraint; carrying out gradient dimensionality reduction by applying sparse random projection to optimize storage calculation; calculating gradient inner product similarity based on the current strategy check point and the verification set to generate an influence score; and finally, iteratively selecting a high-influence data subset according to the score, and optimizing strategy parameters under a course learning framework. According to the method, the data utilization efficiency is remarkably improved, the calculation cost is reduced, and the high-dimensional gradient processing performance is effectively optimized.

The invention discloses a CUDA (Compute Unified Device Architecture) acceleration-based magnetic resonance image rapid reconstruction method and a magnetic resonance imaging system. The method comprises the following steps: acquiring under-sampling k-space data and full-sampling self-calibration signal data; constructing an overdetermined linear equation set; a sparse random projection matrix is designed, and equation set dimension reduction is completed; using CUDA (Compute Unified Device Architecture) to accelerate to solve the linear equation set after dimension reduction, and obtaining a GRAPPA linear combination coefficient; based on the combination coefficient and the under-sampled k-space data, performing parallel synthesis on missing k-space data through CUDA (Compute Unified Device Architecture); and performing two-dimensional Fourier transform and post-processing on the complete k-space data, and outputting a quickly reconstructed 2D magnetic resonance image. According to the method, the dimension reduction advantage of random projection and the parallel computing capability of CUDA are fused, the technical bottlenecks that in traditional 2D GRAPPA reconstruction, the calculation complexity in the calibration stage is high, and the CPU serial processing speed is low are effectively solved, extra hardware transformation or channel compression dependence is avoided, high-quality 2D magnetic resonance images can be stably output, and the clinical rapid diagnosis requirement is accurately met.

The application discloses a data search processing method, device and equipment, and the method comprises the following steps: obtaining a first abstract resilient dataset of a parallel computing framework algorithm according to a dataset of a file system; performing conversion processing on the first abstract resilient dataset to obtain a second abstract resilient dataset; forming a third abstract resilient dataset according to a random projection seed of a random projection tree and the second abstract resilient dataset; performing calculation processing on the third abstract resilient dataset to obtain a fourth abstract resilient dataset; when judging that a leaf node of the random projection tree reaches a condition according to the fourth abstract resilient dataset, obtaining an optimized random projection tree; and searching for target data according to the optimized random projection tree. Through the above method, the application realizes fast retrieval of target data in a high-dimensional space.

This invention discloses a method and apparatus for constructing first-arrival samples based on compressed sensing and unsupervised learning. The method includes: preprocessing raw seismic data; preprocessing includes normalization, flattening, and / or denoising; clustering the preprocessed data using an unsupervised clustering algorithm and principal component analysis to obtain multiple categories of first-arrival sample clusters; for each category of first-arrival sample cluster, generating a random sampling matrix using compressed sensing theory, randomly projecting each first-arrival sample in the cluster to obtain a corresponding low-dimensional compressed representation, and generating random sample numbers based on the low-dimensional compressed representation; randomly selecting first-arrival samples according to a preset ratio based on the random sample numbers, and labeling the selected first-arrival samples to obtain a first-arrival sample set. By integrating clustering and selection, sample construction is achieved. Principal component analysis clustering classification makes the clustering results more stable, and selection after compressed projection ensures uniform coverage and reduces reliance on manual intervention.

The invention provides an anti-collusion verifiable federated learning client selection method, and the method comprises the steps: generating a verifiable public random seed through the cooperation of a client based on a current training round; the client local model updating is completed, dimension reduction is performed on the client local model updating through a public random projection matrix, a model parameter commitment is generated in combination with a private random number, and the model parameter commitment is broadcasted to all participating clients and servers; calculating the selected weight of the client in the current round based on the model quality factor and the long-term fairness factor; based on a zero-knowledge proof system, the verification weight is obtained through calculation according to a preset rule based on committed model parameters; collecting all effective weights, and performing normalization processing to generate effective normalized weights; and based on the public random seed and the effective normalized weight, adopting an unbiased sampling algorithm to select a client.

The application provides a method for risk intelligent early warning of recruitment documents based on feature fusion and black and white lists, comprising: step 1, multi-modal analysis; step 2, marking high-similarity candidate file pairs; step 3, calculating file-level image similarity; step 4, fusing comprehensive similarity; step 5, marking risk levels; step 6, correcting risk levels; and step 7, generating a structured risk report. Through a text fingerprint extraction mechanism based on Burrows-Wheeler transformation and Nilsimsa hash fusion, an image feature extraction mechanism based on random projection and Hellinger distance fusion, and a cross-modal risk comprehensive evaluation mechanism based on multi-modal feature collaborative optimization, the application solves the problems of excessive calculation amount in pairwise comparison of large-scale document sets and multi-modal feature fusion and accurate risk identification.

The application discloses a kind of nonlinear feature extraction methods of medical internet of things data, obtain medical internet of things data and carry out mean removal operation and obtain original data and are divided into training set, verification set and test set;Build suitable for medical internet of things data anomaly detection model, utilize sparse, binary random projection matrix original data is embedded from original feature space to high-dimensional transition space, the dimension of transition space is higher than the dimension of original feature space;Using PCA extracts principal component information and projects data from high-dimensional transition space to low-dimensional feature space to obtain nonlinear projection vector;Using training set and verification set, the best hyperparameter group is obtained by bayesian hyperparameter optimization to anomaly detection model;According to the best hyperparameter group and test set, the performance of the anomaly detection model is calculated;The medical internet of things data to be detected is input into the best final detection model to obtain the corresponding nonlinear feature extraction result.The application realizes very good nonlinear feature processing effect.

The application discloses a kind of lightweight remote sensing image geometric positioning method, system and equipment, the method includes the following steps: S1. Obtain remote sensing image and create coordinate dataset;S2. The parameters of remote sensing image geometric positioning model are initialized, and input weight matrix is generated using orthogonal random projection;S3. Remote sensing image geometric positioning model is trained according to coordinate dataset, hidden layer output is generated by nonlinear feature mapping, and output weight matrix is constructed according to regularized least squares method combined with multi-level solution strategy;S4. According to cross-validation division, the performance of the model is evaluated, and the model is retrained after optimizing the regularization parameter;S5. Based on the model parameters of training completion, test data is input into the trained remote sensing image geometric positioning model to perform regression calculation, and the calculated coordinate data are compared with the coordinates calculated by satellite sensor model, and the running time is recorded, to verify the efficiency and accuracy of the model. The application is applied to the technical field of photogrammetry.

The invention discloses a data selection method for fine tuning of a domain instruction. The method comprises the following steps: randomly sampling from a full-amount data set D to generate a sub-data set; using the basic model to configure LoRA to obtain an agent model Ma0; calculating the LoRA gradient of the total data set D, and projecting the LoRA gradient to a fixed dimension by adopting random projection to obtain an initial gradient vector; finely adjusting the proxy model Ma0 by using the sub-data set to obtain a proxy model Ma1; calculating the LoRA gradient of the total data set D based on the proxy model Ma1, and projecting the LoRA gradient to a fixed dimension by adopting random projection; calculating a gradient distance before and after fine tuning of the full sample; performing descending sorting on the total data set according to the gradient distance, and selecting a front sample to determine a fine tuning data subset; according to the method, the contribution of the sample on the field fine adjustment task is confirmed by using the gradient characteristics before and after fine adjustment through the agent model, the more critical sample for the field task is selected, and the effect that a small-scale sub-data set is superior to a full-scale data set is achieved.

The invention discloses a multi-scale fuzzy uncertainty perception width learning method for ECG classification, and belongs to the technical field of computer-aided medical diagnosis. The method comprises the following steps: carrying out data acquisition and preprocessing on an original electrocardiogram (ECG) signal; constructing a multi-scale feature mapping layer which is used for mapping the preprocessed ECG signals to a plurality of time scales in parallel to obtain high-dimensional feature representation; uncertainty perception enhancement nodes are introduced into the width learning enhancement layer, combined modeling is carried out on input disturbance, structural disturbance and random projection, and enhancement mapping sensitive to uncertainty is constructed; and solving an output weight at an output layer by adopting a fuzzy weighting pseudo-inverse solving mechanism to obtain a discrimination result of the ECG signal category. On the basis of keeping efficient training of a width learning system, the classification robustness, generalization ability and interpretability of electrocardiosignals in a noise environment, form variation and cross-patient difference are remarkably improved, and the method is suitable for real-time diagnosis scenes such as dynamic electrocardiogram monitoring.

The present disclosure relates to an anomaly detection system and method, the system configured to, in an in-situ learning phase or in an inference phase, acquire a data stream; and determine a M*N sketch matrix (Si,j) by using a Count-Min sketching algorithm (CMS) based on hash functions (hi(x)) of hash keys (x) associated with the data stream. Each of the hash functions (hi(x)) comprising a p-stable random projection matrix (Πi) multiplied by the hash keys x, onto a direction of a p-stable random vector (li).

A multi-sensor fusion method and system for identifying rock hardness in coal mines based on geometric consistency graphs is proposed, relating to intelligent coal mining and rock hardness identification during drilling. This method addresses the technical challenges of low accuracy and poor robustness in existing coal mine drilling methods operating under high-noise environments. The method includes: acquiring time-series signals from multiple sensors; introducing pseudo-time coordinates to map the time-series signals into a two-dimensional point cloud; and extracting geometric consistency features of the multi-sensor nodes using random projection; constructing adjacency relationships and weight matrices among the multiple sensors based on geometric consistency metrics; and generating graph data containing topological structure and attribute information by combining the temporal statistical features of the nodes; inputting the constructed graph data into a graph convolutional neural network for feature aggregation and deep learning; and achieving rock hardness category identification through end-to-end training. It is also applicable to rock hardness identification in non-coal mining operations with strong vibrations and complex geological conditions.

The invention provides a heterogeneous analysis federated learning method and system based on a double-anchor mechanism, and belongs to the technical field of artificial intelligence and distributed computing. The server distributes a frozen global backbone network and a random projection matrix to the client, and the client performs projection adaptive processing on local data to obtain enhanced features; updating rigidity is eliminated through a double-anchor-point mechanism: a server aggregates local classifier parameters and an autocorrelation matrix uploaded by each client, iteratively constructs a mask representation matrix and calculates global classifier parameters as global anchor points, and the clients construct double-anchor-point regularization items based on the global anchor points and the local anchor points; and optimizing and solving to obtain a personalized classifier parameter updating local model. According to the method, the problems of feature rigidity and updating rigidity in a data heterogeneous scene are effectively solved, the generalization ability and individuation performance of the model are improved, and the method can be widely applied to privacy-sensitive and data heterogeneous federal learning scenes such as intelligent medical treatment, intelligent finance and intelligent Internet of Things.