19 results about "Conditional mutual information" patented technology

The invention relates to the technical field of data processing, and provides a feature selection-based multi-modal data classification method, which comprises the following steps of: obtaining original data of at least two heterogeneous modals and corresponding initial feature sets; performing intra-modal selection on each modal, constructing a feature association graph based on a graph theory, and screening a core feature subset meeting a threshold requirement through mutual information; constructing a cross-modal feature incidence matrix, realizing inter-modal fusion based on a weighted graph model and condition mutual information, and screening a cross-modal key feature set; and inputting the key features into a classification model to train a multi-modal classifier, and repeating a feature selection process on to-be-classified data to complete classification prediction. According to the method, the optimal threshold value is adaptively determined through innovative fusion of double-stage feature selection, the graph theory and the information theory, the classification accuracy and the data processing efficiency are remarkably improved, and the method can be widely applied to the fields of medical image and pathological report combined diagnosis, automatic driving data fusion, internet multimedia understanding and the like.

The invention discloses an event prediction method and system based on a time sequence hypergraph, and belongs to the technical field of event prediction. The method comprises the following steps: acquiring multi-source heterogeneous data of a target region, defining a unified time index, and generating a time sequence feature sequence of each variable of the region through preprocessing; then, identifying a variable causal relationship in the region based on a frequency domain anti-fact condition mutual information algorithm, and fusing the variable causal relationship with a time sequence evolution relationship to construct a time sequence hypergraph structure representing the interior of the region; a dynamic filter is used for filtering the graph, and deep features of the graph are learned by means of a multi-band spectrum gating mechanism, so that internal complex causal and time sequence modes are effectively captured; and finally, performing dichotomy prediction based on the learned graph representation, and outputting the occurrence probability of future events in the region. According to the method, accurate and explainable event prediction is realized, training and prediction do not need to cross regions, data privacy and calculation efficiency are guaranteed, and stronger robustness is shown for specific data distribution change of the regions.

The invention discloses a preference reinforcement learning-based body-equipped agent strategy training method and system, and the method comprises the steps: obtaining and constructing a multi-source non-stationary environment preference data set: collecting historical interaction trajectories from different bottom-layer dynamic environments, and generating trajectory fragment pairs and corresponding human preference labels; carrying out environment dynamic predictor training, and enabling a calculation control unit to grasp dynamic evolution and dynamic change rules of the body-equipped intelligent body in different physical environments in advance; executing decoupling joint optimization based on conditional mutual information and a regret model; extracting a pure task award decoder which is absolutely immune to any dynamic physical disturbance; security underlying action strategy distillation is executed based on expectation regression, and a final action strategy for controlling physical equipment is safely generated by using pure rewards; and after the strategy network is trained and converged, the processor compiles the strategy network into an executable instruction, and directly deploys the executable instruction to a control unit of the target physical equipment to execute intelligent action control.

The application discloses a drug-resistant gene and resistance category prediction method based on sequence characteristics and a protein language model, and comprises the following steps: 1, constructing an amino acid data set and a drug-resistant gene data set; 2, calculating mutual information, conditional mutual information, Fourier power spectrum characteristics, dipeptide composition, interval amino acid pair composition, mutual information of an ESM2 matrix, interval amino acid pair composition, a tripeptide matrix, a labeled ESM2 matrix and an unlabeled ESM2 matrix; 3, constructing a drug-resistant gene prediction network to obtain a drug-resistant gene prediction score; and 4, constructing a drug-resistant gene resistance category prediction network to obtain a drug-resistant gene resistance category prediction score. The application can realize efficient and accurate prediction of drug-resistant genes and their resistance categories, and the practicability and reliability of the application are verified in the application verification of real genomic data.

The invention discloses a cross-user wearable activity identification method based on group specific concept perception representation learning. The method comprises the steps of collecting and preprocessing multi-user sensor data; the concept offset degree is measured from a time sequence view angle and a semantic view angle, user clustering is carried out by fusing multi-view-angle measurement results, and group specific concept tags are generated; constructing a perception model comprising an activity encoder, a user encoder and a plurality of classifiers, and performing supervised learning by minimizing the joint loss of activity, user and group specific concept classification; introducing a condition discriminator to construct a representation pair of joint distribution and edge distribution, and minimizing condition mutual information of activity representation and user representation under a group specific concept through adversarial training to obtain a trained model; during application, test data are input into the trained model for activity identification. According to the method, through explicit modeling of specific concepts and decoupling of activity and user features, cross-user concept offset is eliminated, and the activity identification generalization ability of the model on new users is significantly improved.

The application discloses a wind turbine early fault warning method and device based on a specific causal network, and relates to the field of data processing, which comprises the following steps: calculating the directed transfer entropy between any two variables in each time window by using conditional mutual information according to the data matrix of each time window, constructing a specific causal network of each time window and calculating the variation causal feature of the current time window; calculating the causal variation potential dynamic network marker score of each node in the specific causal network of each time window and the comprehensive state index value of each time window according to the variation causal feature of each time window, and calculating the change value of the comprehensive state index value of the current time window and the previous time window; determining whether the current time window is a warning window based on the change value of the comprehensive state index value. The application solves the problem that it is difficult to perform early warning on SCADA data.

The invention discloses a method for analyzing an influence value of operating parameters of a coal-fired unit on power supply coal consumption. The method comprises the following steps: acquiring operation parameter data and a power supply coal consumption value from a coal-fired unit DCS (Distributed Control System), and preprocessing and discretizing; a mutual information matrix M belongs to the set is constructed, and conditional mutual information I (X; x [delta] C) quantizes the coupling relationship between the parameters; establishing a three-order tensor T which belongs to a characteristic'parameter-parameter-coal consumption 'nonlinear coupling relation; carrying out Tucker decomposition on the tensor; and calculating an independent influence item and a coupling influence item of the parameter based on a decomposition result to realize quantitative separation of the coupling effect. The method further comprises a dynamic updating mechanism, incremental updating is carried out every 8 hours, and working condition changes are rapidly adapted through combination of CP decomposition and self-adaptive weight adjustment. According to the method, the defect that the parameter coupling effect is ignored in a traditional method is overcome, the coupling effect recognition accuracy is improved by 35-40%, the power supply coal consumption prediction error is reduced to + / -0.8 g / (kW.h), and a reliable decision basis is provided for optimized operation of a unit.

The invention discloses a social network-oriented privacy ontology construction method and system, and relates to the technical field of security information processing, and the method comprises the steps: obtaining a social network data set, and extracting candidate privacy attributes; calculating information entropies of the candidate privacy attributes and mutual information of the candidate privacy attributes and identity or target sensitive attributes, normalizing the information entropies, and weighting the information entropies and the identity or target sensitive attributes to obtain a comprehensive sensitivity score so as to screen a privacy attribute set; conditional mutual information of the privacy attribute pair is calculated under the constraint of conditional attributes to serve as dependency intensity, an inference direction is determined based on a conditional entropy reduction criterion and normalized to obtain a relation weight, and a privacy ontology containing attribute nodes and weighted dependency edges is constructed; and receiving data batch according to time slices, executing node addition detection, statistic increment updating and relation evidence accumulation, and incrementally updating an ontology structure or weight. Through the technical scheme of the invention, privacy attribute sensitivity quantification and inference relation modeling are realized, and privacy risk identification accuracy and interpretability are improved.

The invention discloses a clinker free calcium prediction method based on causal alignment, and relates to the technical field of intelligent control of a cement production process. The method comprises the following steps: firstly, carrying out multi-source data acquisition and fusion to obtain process variables and test data; the method comprises the following core steps of: performing refined time alignment on variables based on process time delay grouping so as to reflect differential delay of different working sections on clinker quality; online causal filtering is carried out by adopting a unilateral Gaussian kernel, and future information leakage is strictly avoided while noise is eliminated; carrying out causal feature extraction and screening by using conditional mutual information and a minimum redundancy maximum correlation algorithm under the constraint of a structural causal model and time sequence precedence; and finally, generating an aggregation statistical characteristic sample in the causal time window for constructing a prediction model and outputting a real-time prediction value of free calcium. According to the method, the problems of low prediction precision, poor interpretability and difficulty in cross-line migration caused by rough time delay processing, data noise and pseudo correlation in the prior art are effectively solved.

The utility model discloses a kind of utility enhanced conditional characteristic selection differential privacy data publishing method, the method is first to the original data set is preprocessed, then utilize conditional mutual information to the data set after pre-processing is associated with feature selection;Again normalization data, the data value corresponding to the feature that meets threshold condition is micro-aggregated and handled, obtain several cluster of the size k of scale, and using profile coefficient calculation obtains the local optimum k value;Then, according to the redefined feature dependence sensitivity, add the noise that meets condition to each cluster, reassign privacy budget to realize differential privacy;Finally, for the data after disturbance is published, the data published can be counted and classified etc. Task analysis is analyzed. The utility model can resist individual data privacy attack of enemy with strong background knowledge, improve the availability of data publishing under the premise that individual sensitive data is guaranteed privacy.

The invention provides a data transmission analysis method, device and equipment based on multi-scale condition mutual information, and relates to the technical field of data mining. The method comprises the following steps: performing Morlet wavelet decomposition on an original sequence to obtain wavelet coefficients under a plurality of time scales, and constructing a candidate lag variable set corresponding to the time scale based on each wavelet coefficient under each time scale; screening a key lag component in each candidate lag variable set by adopting a conditional mutual information progressive strategy to obtain an optimal embedding vector corresponding to each time scale; and for each time scale, estimating conditional probability density and marginal probability density based on the corresponding optimal embedding vector, calculating conditional mutual information, and substituting the conditional mutual information into a corresponding variable embedding data transmission intensity calculation formula to obtain data transmission intensity and direction between the first original sequence and the second original sequence. According to the method, multi-scale, self-adaptive and high-precision depiction of the dynamic influence relationship among the cross-stage engineering data can be realized.

The application discloses a cross-user wearable activity recognition method based on group-specific concept-aware representation learning, which comprises the following steps: collecting multi-user sensor data and preprocessing; measuring the concept drift degree from the time sequence perspective and the semantic perspective respectively, fusing the multi-perspective measurement results to perform user clustering, and generating group-specific concept labels; constructing a perception model comprising an activity encoder, a user encoder and multiple classifiers, and performing supervised learning by minimizing the joint loss of activity, user and group-specific concept classification; introducing a conditional discriminator to construct a representation pair of joint distribution and marginal distribution, minimizing the conditional mutual information of activity representation and user representation under the group-specific concept through adversarial training, and obtaining a trained model; and inputting test data into the trained model for activity recognition. The application explicitly models the group-specific concept and decouples the activity and user features, eliminates the cross-user concept drift, and significantly improves the activity recognition generalization ability of the model on new users.

The application discloses a data governance effect evaluation method, and relates to the technical field of data governance. Firstly, the application collects evaluation index time series of a complete business cycle before and after the implementation of a governance action, and then, after discretization processing, the application screens the correlation edges between indexes based on mutual information and marks the correlation polarity, so as to construct an undirected dependence network before and after the governance. Then, the application uses conditional mutual information to complete the discrimination of the causal direction, and generates a corresponding causal graph. Finally, the application compares the edge changes of the causal graph, identifies the effect offset phenomenon between indexes, calculates the net effect value, and realizes threshold early warning. The application can accurately capture the index causal dependence structure changes caused by the governance, and improves the accuracy and reliability of the data governance effect evaluation.

The application provides a multi-level causal inference method and equipment for multi-omics heterogeneous data. By establishing a feature-level causal constraint based on conditional mutual information and designing a two-stage causal inference scheme of feature level and representation level, the multi-level causal relationship is systematically inferred from the multi-omics heterogeneous data. The essence of the multi-omics causal relationship is strictly captured by using the conditional mutual information. The framework establishes a strict mathematical basis, so that the causal inference is improved from experience to theory, and the risk of false causal discovery is greatly reduced. The two stages of feature level and representation level are mutually constrained and supplemented. The feature level constraint ensures the accuracy of the basic causal relationship, and the representation level inference captures complex multi-step causal chains. Compared with the single-level method, the multi-level design can capture more rich causal structures.

The application relates to a personalized federated learning method, device and system based on global feature sharing. The personalized federated learning method based on global feature sharing is applied to a client and comprises the following steps: receiving a global feature extractor model and global features sent by a server; initializing a local model according to the global feature extractor model and a local classifier model; inputting local image data into the initialized local model for model training, determining a loss function of the local model, wherein the loss function comprises a cross-entropy loss between a training label and an actual label of the local image data and a conditional mutual information regularization term; performing first updating processing on the local model based on back propagation according to the loss function of the local model; and determining a target local model when the local model converges. By introducing the global features and the conditional mutual information regularization term, the global features are shared, the generalization performance of the local model is improved, and overfitting of the local model is prevented.

The invention discloses an APT attack detection method and device based on traceability graph subgraph division of mutual information approximation, equipment and a medium, and relates to the technical field of network security, and the method comprises the steps: constructing an original traceability graph comprising node attributes and edge attributes; performing representation learning by using a graph convolutional neural network model to generate a low-dimensional node embedding vector; on the basis of a conditional scoring function model, a node embedding vector is used as input, and in combination with context conditions, conditional mutual information between nodes is approximately calculated; the context condition is determined according to the edge attribute; based on the condition mutual information between the nodes, an optimization objective function model is constructed, and sub-graph division of the original traceability graph is completed; sub-graphs obtained through division are mapped to an existing attack knowledge framework to be labeled, key nodes are recognized by calculating the contribution degree of nodes to mutual information in the sub-graphs, a path formed by the key nodes is recognized as a key attack chain, and a visualization result and alarm information are generated. According to the method, the calculation efficiency, the robustness and the interpretability are improved.

This invention discloses a method and device for early fault warning of wind turbines based on specific causal networks, relating to the field of data processing. The method includes: calculating the directed transfer entropy between any two variables in each time window using conditional mutual information based on the data matrix of each time window; constructing a specific causal network for each time window and calculating the variational causal characteristics of the current time window; calculating the dynamic network marker score of the causal variation potential of each node in the specific causal network of each time window and the comprehensive state index value of each time window based on the variational causal characteristics of each time window, and calculating the change in the comprehensive state index value between the current time window and the previous time window; and determining whether the current time window is a warning window based on the change in the comprehensive state index value. This invention solves the problem of difficulty in providing early warnings based on SCADA data.

The invention provides a commodity association rule intelligent analysis method and system based on knowledge mining, and relates to the field of data mining, and the method comprises the steps: carrying out the semantic alignment of a commodity object set and a commodity knowledge base, and constructing a multi-dimensional feature tensor; carrying out iterative optimization on the tensor by utilizing a constraint operator of semantic constraint relationship conversion between entities to generate a hierarchical cluster structure; counting co-occurrence frequency of commodities in the cluster, calculating a condition mutual information matrix, identifying a cluster pair corresponding to an abnormal peak value, and modeling into a directed causal graph; calculating path intensity variation through anti-fact intervention, screening stable path edges, and converting the stable path edges into association rules; mapping the association rules to a knowledge base search reasoning chain, and eliminating low-quality rules based on isomorphic metric values to obtain a verification rule set; and finally calculating posterior probability distribution in combination with the target commodity and outputting an associated commodity. According to the method, deep semantic association can be mined, the accuracy and interpretability of association rules are improved, and false association is effectively avoided.

A system, method, and computer program product for training a deep neural network. The deep neural network may be trained using a learning process defined to simultaneously optimize an error function of the deep neural network and a network mapping function of the deep neural network. The network mapping function may represent predicted tag distribution geometries for the deep neural network. The learning process can improve the accuracy of the trained deep neural network model and the robustness of the trained deep neural network model against adversarial attacks. Optimizing the network mapping function may also provide deeper knowledge of the trained deep neural network model operation, which may facilitate an interpretability enhancement of the trained model, thereby encouraging the popularization and application of the trained model.