18 results about "Dirichlet distribution" patented technology

The invention discloses a credible medical image segmentation method based on non-independent evidence fusion and uncertainty decoupling, and aims to solve the problem of evidence explosion caused by over-confidence and non-independent evidence in a medical high-risk scene. According to the method, improved UNet + + is utilized to extract spatial and semantic view angle features, and the spatial and semantic view angle features are mapped into Dirichlet distribution parameters; a discount factor is generated through recursive discount fusion, attenuation processing is performed on incremental belief, and active evidence fusion is realized; explicitly decoupling into data and model uncertainty under a second-order probability framework based on a Bayesian variance decomposition theory, optimizing segmentation loss by using the data uncertainty, and guiding redundant discount by using the model uncertainty; a discount regular term is introduced, so that the model is prevented from being forgotten while redundant evidences are reduced; and constructing a joint optimization objective function, training the model in stages, and outputting a segmentation result and a pixel-level uncertainty evaluation result. According to the method, uncertainty evaluation is provided while high-precision segmentation is kept, and the safety of clinical application is remarkably improved.

The invention relates to a multi-modal data fusion method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: respectively carrying out feature extraction on image input data from at least two modes to obtain an image feature vector of each mode; on the basis of the image feature vector of each mode, Dirichlet distribution parameters corresponding to each mode are generated, the distribution difference between the modes is evaluated, and the Dirichlet distribution parameters are used for representing the evidence intensity of each category in the classification or clustering task; according to the distribution difference, fusing the Dirichlet distribution parameters of each mode to obtain a basic probability distribution result after fusion; and dynamically adjusting the basic probability distribution result by using a Kalman filter, and outputting a final classification or clustering result. By adopting the method, the fusion weight of each modal data can be dynamically adjusted, and the accuracy and reliability of a multi-modal fusion system are improved.

Apparatuses, methods, and systems for determining winning arms of electronic testing. One method includes obtaining historical data values related to the A / B test of a user, storing the historical data values, determining a historical weight for the historical data values, receiving new data values from the plurality of computing devices collected based on recipient actions during execution of the A / B, constructing a Dirichlet distribution, inferring corresponding central tendencies of samplings of a metric distribution, wherein each central tendency of the corresponding central tendencies is determined by sampling the Dirichlet distribution, constructing an overall utility distribution for each arms of the A / B test by combining the central tendency of each sampling of the metric distribution with a corresponding sampling of a conversion probability distribution, determining a winning arm of the A / B testing by comparing the overall utility distribution of each arm with each other arm of the A / B test.

The application provides a threat intelligence quality crowdsourcing evaluation method based on a reputation mechanism, and relates to the field of computer network security.The method provided by the application comprises the following steps: generating an objective score based on objective dimension evaluation by an automatic detection technology; obtaining an initial reputation value based on an additional reputation value and an initial reputation contribution value, and obtaining an initial subjective score of a subjective dimension by a crowdsourcing evaluator; screening out non-anomalous scores based on a weighted absolute median deviation, and obtaining a subjective score and a score entropy by post-hoc aggregation of the non-anomalous scores based on a weighted Dirichlet distribution model; obtaining a score deviation based on the initial subjective score and the subjective score, and constructing a reputation function to update the initial reputation value to obtain a reputation value; constructing a reward function based on the score deviation and the reputation value, and distributing rewards to the crowdsourcing evaluators; and obtaining a quality score by weighted fusion of each score.The application effectively realizes quality evaluation of threat intelligence by combining a crowdsourcing mechanism and a reputation reward mechanism.

The invention discloses a text-image pedestrian re-identification method oriented to semantic fuzziness and excessive decision confidence, and belongs to the technical field of pedestrian re-identification. The method mainly comprises the steps that based on a distributed feature alignment module, image and text modal features are coded into Gaussian distribution, a feature center and an intra-class semantic change range are jointly expressed through a mean value and a variance, and the modeling ability of cross-modal semantic difference and visual diversity is improved; the uncertainty penalty alignment module based on evidence deep learning explicitly models the uncertainty in the image-text matching process by introducing Dirichlet distribution and subjective logic theories, and applies adaptive penalty to a high-uncertainty matching relationship, so that the excessive confidence of the model on a mismatching result is inhibited. According to the method, a probability distribution-based feature expression mode and an uncertainty perception training mechanism are adopted, so that the matching accuracy and generalization ability of a cross-modal retrieval system in view angle change, semantic fuzziness and background interference scenes are remarkably improved.

The present application relates to a pathological whole slice image classification method based on multi-branch independent mask and Dirichlet evidence fusion, belonging to the cross field of biological information and artificial intelligence. In view of the defects of traditional multi-instance learning method in weakly supervised classification task of pathological whole slice image, such as excessive attention concentration and static fusion, the present application sets dynamic mask parameters through multi-branch independent setting, forces different branches to pay attention to different pathological regions, and solves the problem of insufficient feature diversity caused by attention concentration; combining the confidence and uncertainty of Dirichlet distribution quantization branch prediction, the branch fusion weight is dynamically adjusted based on evidence theory, and the fusion robustness of multi-branch prediction result is improved. The experiment is verified on the public pathological data set such as CAMELYON-16, compared with the MIL method, the present application improves the AUC index by 1.1-2.4%, and significantly enhances the accuracy and generalization ability of pathological WSI classification.

A device and method for reviewing literature by using Latent Dirichlet Allocation (LDA) is proposed. The device may include a pre-processing unit extracting text data for modeling, and a modeling unit automatically classifying topics as many as a set number (K) and generating a probability distribution of the topics by literature and a probability distribution for words by topic. The device may also include a clustering unit updating the number (K) of the topics, an interest analysis unit confirming trends by topic over time, and a generality analysis unit quantitatively confirming a research scope of each specific topic. The device may further include a similarity analysis unit quantitatively confirming a similarity between the topics, a network analysis unit quantitatively confirming a correlation between the topics, and a display displaying the trends by topic over time, research scope of each specific topic, similarity between the topics, and correlation between the topics.

The application discloses a marine abnormal target discrimination method based on frequency analysis and grid adaptive expansion. A radar detection area is divided into grids with equal longitude and latitude span, and target frequency data in each grid is counted based on historical AIS track data of each type of target. Then, historical speed and acceleration of different types of targets are counted to obtain target normal activity representation expressed by a Gaussian probability density distribution function. A plurality of random samples are obtained from Dirichlet distribution, and target frequency data statistical results are recalculated in sequence to quantize uncertainty of the target normal activity mode. A cross-subspace support area of an arbitrary grid is formed through an adaptive expansion mechanism based on grid length and uncertainty similarity, and the target activity path coincidence average is calculated based on the uncertainty quantization result of the target type normal activity mode and the cross-subspace support area of the arbitrary grid, and the abnormal target is judged by combining the target speed and other feature coincidence calculation.

This application provides a UAV collaborative management method based on the fusion of spatiotemporal uncertainty closed-loop perception and dynamic credibility, including: S1: using evidence deep learning to perform second-order probability modeling on the collected multidimensional heterogeneous data, and using Dirichlet distribution modeling for classification tasks to obtain a second-order probability feature map containing data uncertainty and model uncertainty, and encapsulating the positioning information, classification information, and the two types of uncertainty into node attributes of a dynamic semantic graph; extracting the target's positioning and classification information, and fusing the dual uncertainties through a dynamic weight adjustment mechanism based on environmental parameters, positioning accuracy, and historical performance to generate a comprehensive uncertainty measure; in a multi-UAV collaborative field... In this scenario, the credibility consistency of the perception results of each UAV is verified. Spatial consistency index, temporal jump penalty factor and historical reputation score are calculated and fused to generate credibility consistency verification index. The fused credibility is mapped to airspace risk level. In the multi-UAV collaborative scenario, a potential game loss function is constructed with risk value as constraint, Nash equilibrium is solved and collaborative avoidance trajectory is generated, where the virtual collision radius is dynamically scaled according to the risk value. The real trajectory and predicted trajectory returned by the execution layer are obtained and the residual is calculated. When overconfidence of the perception model is detected, feedback gradient is generated through the edge lightweight large language model to correct the relevant parameters of the evidence deep learning network.

The invention relates to the technical field of human deep neural network model quantification, and discloses a hierarchical variance propagation and gradient cooperative modulation-based neural network uncertainty quantification method, which comprises the following steps: S1, parameter posteriori distribution extraction: using a Bayesian neural network trained by Gaussian prior with a parameter of the mean value to obtain posteriori distribution with the same parameter, and extracting the posteriori distribution; wherein the parameter mean value is a variance; s2, posterior distribution truncation: truncating posterior distribution by using a dynamic distribution truncation mechanism; s3, constructing a continuous belief function, carrying out Mobius inversion, and obtaining confidence quality distribution for uncertainty expression from the belief function; and S4, fitting Dirichlet distribution, and constructing a hierarchical variance propagation matrix. And S5, carrying out further quantification on the trained model by adopting gradient weighting, layer selection gradient and gradient perturbation integral methods of a specific category. According to the method, the reliability, stability and risk controllability of model prediction are remarkably improved.

The application relates to an event prediction method based on pair event relation learning and multi-view evidence fusion, and belongs to the technical field of natural language processing. The method comprises the following steps: pair event relation learning: a pre-training language model is trained by constructing positive and negative samples in a supervised learning framework; an event relation graph is constructed: the pre-training language model after pair event relation learning is used to deduce the dependency relation of the pair event, the probability output by the model is used as the weight of the edge, and thus the event relation graph is constructed; multi-view evidence learning: quantitative evidence of a candidate event is learned from a text semantic view and a graph structure view; credible evidence fusion: the uncertainty of each view is modeled by using a Dirichlet distribution, and multi-view evidence is dynamically fused by using Dempster-Shafer theory to generate a final prediction result. Under the condition of not depending on an external knowledge base, the application realizes a prediction accuracy of 64.62% on an NYT data set, and exceeds an existing baseline model.

The present application belongs to the technical field of medical data analysis and artificial intelligence auxiliary diagnosis, and provides a medical diagnosis prediction method and system based on knowledge guidance and credible evidence learning. Electronic medical record data of a patient to be tested is acquired; a trained credible diagnosis evolution and prediction model is used to perform diagnosis prediction on the electronic medical record data of the patient to be tested, and a high-risk blind area early warning is triggered when the uncertainty score is higher than a set threshold; the training process of the credible diagnosis evolution and prediction model is as follows: a global representation vector of the patient is generated, and then positive evidence and negative evidence of various diseases are obtained and quantified as Dirichlet distribution, and prediction probability and prediction uncertainty score of various diseases are calculated; a binary evidence loss function is used to calculate prediction error, and model parameters are optimized. It can not only provide high-precision diagnosis prediction results, but also accurately quantify cognitive uncertainty.

The invention belongs to the technical field of image information processing, and discloses an image generation quality evaluation method for semantic evidence learning. Constructing a semantic association extraction network, extracting local features of the image and the text by using a CLIP pre-training model, and constructing a semantic association matrix according to the local features to quantify association strength among all local feature pairs; and aggregating the matrix into a semantic intensity vector through pooling operation. Constructing a semantic evidence learning network, mapping a semantic intensity vector into an evidence vector, parameterizing Dirichlet distribution according to the evidence vector, and modeling distribution of association intensity into probability distribution on a predefined quality level; the image quality score is directly derived by calculating the expected value of the distribution. According to the method, fine-grained correlation quantification can be carried out on the embedding characteristics of the image and the text, and the intensity distribution is revealed; and the two-way evaluation evidence is subjected to probabilistic modeling and fusion through Dirichlet distribution, so that the interpretability, the robustness and the discrimination accuracy of quality evaluation are remarkably improved.

The application belongs to the technical field of image information processing, and discloses a semantic evidence learning image generation quality evaluation method. A semantic correlation extraction network is constructed, a CLIP pre-training model is used to extract local features of images and texts, and a semantic correlation matrix is constructed according to the local features to quantify the correlation strength between all local feature pairs. The matrix is aggregated into a semantic strength vector through a pooling operation. A semantic evidence learning network is constructed, the semantic strength vector is mapped into an evidence vector, and the correlation strength distribution is modeled as a probability distribution on a predefined quality level according to the parameterized Dirichlet distribution. The image quality score is directly derived by calculating the expected value of the distribution. The method can not only quantize the fine-grained correlation of the embedded features of images and texts and reveal the strength distribution, but also probabilistically model and fuse the bidirectional evaluation evidence through the Dirichlet distribution, which significantly improves the interpretability, robustness and discriminant accuracy of the quality evaluation.

The invention discloses a radar interference effect evaluation method based on a multi-level subjective and objective information fusion Bayesian network, which is applied to the crossing field of radar signal processing and electronic countermeasure, and aims at the limitation of the existing radar interference effect evaluation method, the method comprises the following steps of: 1, constructing an evaluation index set; 2, proposing a subjective and objective fused adaptive weight calculation model; 3, constructing a directed acyclic graph by taking the evaluation index in the step 1 as a father node and the interference effect level as a child node so as to design and obtain a Bayesian network topological structure; fourthly, expert knowledge and the self-adaptive weight in the second step are fused through an interference effect grade-evaluation index-state knowledge matrix, and a three-dimensional conditional probability table is generated; 5, constructing parameter samples before and after interference; and 6, taking Dirichlet distribution as prior distribution of conditional probability, determining network hyper-parameters according to the initial conditional probability obtained in the step 4, training the network based on the parameter samples in the step 5, and finally performing interference assessment based on the trained network.

Apparatuses, methods, and systems for determining winning arms of electronic testing. One method includes obtaining historical data values related to the A / B test of a user, storing the historical data values, determining a historical weight for the historical data values, receiving new data values from the plurality of computing devices collected based on recipient actions during execution of the A / B, constructing a Dirichlet distribution, inferring corresponding central tendencies of samplings of a metric distribution, wherein each central tendency of the corresponding central tendencies is determined by sampling the Dirichlet distribution, constructing an overall utility distribution for each arms of the A / B test by combining the central tendency of each sampling of the metric distribution with a corresponding sampling of a conversion probability distribution, determining a winning arm of the A / B testing by comparing the overall utility distribution of each arm with each other arm of the A / B test.

The invention discloses a class imbalance data oversampling method based on Gaussian process regression, and belongs to the field of machine learning data preprocessing. Aiming at the problems that an existing oversampling technology (such as SMOTE) neglects feature correlation, sample generation is unreasonable and scene adaptability is poor, the method comprises the steps of identifying minority class samples and calculating feature boundaries, constructing a k-nearest neighbor model to obtain a local sample set, training a GPR model for each feature dimension to capture feature correlation, generating input points based on Dirichlet distribution or random indexes, and obtaining a local sample set. Adaptive noise is added in combination with GPR prediction uncertainty, samples are cut to a reasonable range, non-repeated samples are screened, and finally a balanced data set is generated through combination. The method supports multi-dimensional and single-feature scenes, automatically balances category distribution, combines samples to fit real data distribution, can effectively improve the model performance of category imbalance classification tasks, and is suitable for financial risk control, medical diagnosis and other scenes requiring accurate identification of minority class events.

The application discloses a dynamic data-driven risk control method for health insurance claim settlement, comprising: collecting and preprocessing claim text data to generate a text corpus; screening normal claim settlement samples based on compliance labels, latent Dirichlet distribution topic modeling, and constructing a topic probability feature library; K nearest neighbor algorithm comparison to determine the K nearest normal claim settlement samples; calculating an abnormal score, comparing with a preset threshold, and identifying and warning abnormalities; dynamically collecting and updating data, regularly adjusting parameters, and forming a risk control closed loop. The application realizes automatic abnormal detection and risk warning for the whole process of health insurance claim settlement, and improves the intelligent and accurate level of claim settlement risk control.