9 results about "Nearest neighbor clustering" patented technology

A collective data leakage attacks on a Retrieval-Augmented Generation (RAG) application for a generative artificial intelligence (GenAI) application is detected and prevented based on an analysis of incoming queries to distinguish normal and potentially malicious querying behavior. Similarity distances associated with each query are determined, such as the distances between the query vector embeddings and the nearest neighbors in the vector embedding space, distances between each query vector embedding and other query vector embeddings for other queries from the same user, or distances between each query vector embedding and the nearest neighbor cluster of vector embeddings in the vector space. A data leakage attack on the RAG application may be determined based on the similarity distances associated with each query. In response to the identification of a potential data leakage attack, the release of data from the RAG application may be halted.

The invention discloses a neighbor clustering method and device for complex manifold data, and relates to the technical field of data mining. The method comprises the following steps: firstly, obtaining lambda-nearest neighbors of each data object by using a natural neighbor search algorithm, and calculating the density; then determining a natural density peak, and dividing other objects into sub-clusters to which the natural density peak belongs according to representative information of the other objects; setting a density threshold value tau, determining a non-noise data object, and performing clustering by using a k-nearest neighbor of the non-noise data object; and finally, dividing the noise data object into the class cluster to which the natural density peak belongs, thereby finishing the clustering analysis of the whole data set. The invention discloses a neighbor clustering method for complex manifold data, which eliminates the interference of noise points in the clustering process, performs fast clustering by utilizing neighbor information, can accurately identify any shape class clusters in the complex manifold data, has excellent adaptability and noise resistance to high-noise data, and can be used for high-precision clustering of the complex manifold data. And obvious high efficiency and robustness are shown.

This invention discloses a spatiotemporal clustering method for small target detection based on sparse event points. The method involves preprocessing raw data captured by an event camera to obtain preprocessed event data; traversing all event points in the event data, assigning a weight value to each event point, and sorting all event points in descending order according to their weight values; using the average weight of the bottom M% of event points as a clustering threshold, and performing nearest neighbor clustering on the event data based on this threshold to obtain preliminary detection results; and finally, performing point cloud filtering on the preliminary detection results to obtain the final detection results. This invention can directly extract small target features from event data for detection. Under static backgrounds, it can effectively filter out event point clutter triggered by thermal noise and interference event points generated by the movement of large objects, achieving high detection accuracy.

The application discloses a kind of adaptive learning method and system for underground cable breakage risk assessment, it is related to cable breakage risk assessment field;The method comprises: constructing the state evolution vector of M cable breakage events, the first similarity of K state evolution clusters and real-time evolution vector is determined to the near neighbor clustering of M state evolution vectors, generates K state evolution cluster, according to K first similarity, constructs similar cluster sequence, in similar cluster sequence, the variance of several relative time area numbers in Q similar clusters is calculated, according to the variance of several relative time area numbers in Q similar clusters, determine adaptive learning cluster, determine the second similarity in adaptive learning cluster, and calculate breakage risk index based on second similarity;The application generates breakage risk index based on the weighted relationship of the L2 norm of each historical state evolution vector in adaptive learning cluster and second similarity, realizes the intuitive quantification of breakage risk assessment.

The invention provides a different-vehicle-type electric vehicle path planning method and device based on a multi-strategy fusion Jaya algorithm. The method comprises the following steps: acquiring vehicle information and node information; constructing a fitness function based on the total path cost function; generating an initial population by using a nearest neighbor clustering method according to the vehicle information and the node information; for the current population, a double-elitist adaptive reservation strategy, an updating strategy of a Jaya algorithm, a Monte Carlo acceptance criterion and a vehicle type adjustment strategy are comprehensively adopted to carry out iterative optimization, a greedy algorithm strategy is used to insert charging station nodes, and a reverse energy calculation strategy is used to optimize the positions of the charging station nodes; obtaining a new population Dpop through the iterative optimization process; and outputting a global optimal solution in the new population Dpop.

The application provides a big data feature portrait generation method based on multi-stage optimization, a medium and a system, and belongs to the technical field of feature engineering. According to the application, the information bottleneck compression model is used to evaluate feature importance, the Bayesian multiple imputation method is used to fill in missing values for high-value features, the robust Z-score standardization is performed on unified wide tables, and the wavelet multi-scale decomposition stabilization processing is performed on non-stationary time series features. The number of principal components is determined through progressive sampling and parallel analysis, a projection matrix is constructed to realize dimension reduction, the local adaptive Mahalanobis distance and shared nearest neighbor clustering are used in a low-dimensional subspace, the cluster feature portrait is extracted by backtracking to the original feature space, and when the density is uneven, hierarchical adaptive subspace division is performed to optimize clustering. The technical problem that the clustering result accuracy is insufficient due to uneven data quality in the feature portrait generation process of multi-source heterogeneous data in a big data scene is solved.

The embodiment of the invention discloses a vector data learning type indexing method and device, equipment and a storage medium, and relates to the technical field of artificial intelligence and databases, the method comprises construction operation, retrieval operation and updating operation of an index structure, and the index structure is composed of a mapping relation table from cluster numbers to vector ID sets and a classification model. According to the construction operation, reference vectors in a vector data management system are divided, a mapping relation table from cluster numbers to vector ID sets is formed, and meanwhile, a classification model is trained to be used for predicting L-nearest neighbor clusters of query vectors. In the retrieval operation, candidate vectors are collected through one-time prediction of the classification model, and k designated candidate vectors with the closest distance are returned by calculating the preset distance between the query vector and the candidate vectors. The updating operation updates the index structure in real time according to addition, deletion, modification and change of the reference vector. The invention provides a construction method and a dynamic updating mechanism of a vector data learning type index structure, and more efficient approximate nearest neighbor search is realized.

The application provides a seismic attribute fusion method based on a thin-plate spline base function, comprising the following steps: step 1, calculating an attribute value set X of all seismic attributes participating in fusion; step 2, defining a thin-plate spline base function; step 3, establishing an initial network; step 4, calculating the distance of sample data to a cluster center; step 5, calculating the nearest neighbor cluster of the kth sample data x k ; step 6, forming an activation function of the thin-plate spline base function; step 7, establishing a network learning error judgment function; step 8, correcting a training weight coefficient, and completing seismic attribute fusion calculation according to the trained weight coefficient w i . The seismic attribute fusion method based on the thin-plate spline base function is more consistent with geological characteristics, meets the requirement of nonlinear seismic attribute fusion on a mathematical model, solves the nonlinear relationship fusion processing of seismic attribute data, and can enhance the accuracy of seismic attributes in reservoir prediction.

The application discloses a kind of based on nearest neighbor clustering forecast rolling mill machine hour output method, comprising: obtaining the real-time rolling data of each material;Based on rolling time, the interval time, rhythm time and machine hour output of each material are calculated, and machine hour output information table is established;Based on material steel grade, material weight and product specification setting classification condition, using nearest neighbor clustering algorithm based on classification condition real-time to material is classified, and rolling mill machine hour output grading table is established;Real-time update each kind of product specification rolling mill interval time, rhythm time and machine hour output data and establish rolling mill machine hour output gear information log table;Obtain the information of material to be rolled, and based on each table, unknown rolling mill machine hour output is predicted.The application method is according to the characteristics of many rolling mill rolling product specifications, using the method of nearest neighbor clustering to integrate and cluster within the range of difference allowed specification, facilitate production forecast and production arrangement etc.Work is carried out, effectively improve production efficiency.