33 results about "Multiview learning" patented technology

A system and method for detecting anomalies in a communication network includes detecting first outliers in a first set of quality indicators for a cellular group, detecting second outliers in a second set of performance indicators for the cellular group, correlating the first outliers and the second outliers to produce an anomaly candidate, determining a confidence threshold for the anomaly candidate, and indicating a network anomaly in response to the confidence threshold exceeding a predetermined threshold.

A system and method for detecting anomalies in a communication network includes detecting first outliers in a first set of quality indicators for a cellular group, detecting second outliers in a second set of performance indicators for the cellular group, correlating the first outliers and the second outliers to produce an anomaly candidate, determining a confidence threshold for the anomaly candidate, and indicating a network anomaly in response to the confidence threshold exceeding a predetermined threshold.

The invention relates to a search engine user information demand satisfaction evaluation method capable of integrating multiple views and semi-supervised learning. The method is divided into the following six stages: preprocessing data, training a subview satisfaction model, distributing a dummy tag for unlabeled data, training a user satisfaction model based on the multiple views and the semi-supervised learning, and carrying out evaluation. Through a semi-supervised learning method, a small quantity of labeled data and a great quantity of unlabeled data are used for improving the performance of an evaluation model, and a multi-view learning method is imported to overcome the problem that a traditional single-view based semi-supervised learning method is always caught in local optimum. The search engine user information demand satisfaction evaluation method has the beneficial effects: (1) under the condition of the small quantity of labeled data, the search engine user information demand satisfaction can be effectively evaluated; (2) the small quantity of labeled data and the great quantity of unlabeled data can be used for improving the evaluation performance of the user satisfaction model; and (3) a search process of the user can be independently described from angles of behaviors and time, and the model can be prevented from being caught into the local optimum through mutual learning.

The invention discloses a city water logging analysis method based on transfer learning. The method comprises steps as follows: firstly, an area is divided according to characteristics of a city; then, features are constructed by use of social media in each area and sensor data, and the features are fused to the best by use of multi-view learning; finally, big city water logging knowledge with the rich data volume is transferred to smaller cities through transfer learning, and a water logging severity degree model for the smaller cities is trained. Different view data are fused through multi-view learning, data from other sources can be effectively extended, and meanwhile, on the basis of a transfer learning technology, the method can also have a better effect for small cities with smaller samples.

The invention discloses a multi-view learning method and system for rhesus monkey eye movement decision decoding, and belongs to the field of multi-view decoding in intrusive brain-computer interfaces. The multi-view learning method comprises the steps that an eye movement decision decoding model comprising a characteristic node extraction network, a reinforced node extraction network and a predicting network is established; eye movement decision direction data obtained after encoding of the local field potential and action potential of a medial frontal cortex auxiliary eye area in a rhesus monkey intrusive brain-computer interface, and one-hot encoding are input into the eye movement decision decoding model for training, and thus the trained eye movement decision decoding model is obtained; and the to-be-decoded local field potential and action potential are input into the trained decoding model, and thus a decoding result of the eye movement direction is obtained. The multi-view learning method and system are less in using limitation, low in computation complexity and suitable for decoding of action potential and local field potential signals in the intrusive brain-computer interface and other various multi-view learning scenes.

The invention discloses a fraud marking method and device based on multi-view learning, and the method comprises the steps: obtaining a user data set; dividing the user data set into a marked data setand an unmarked data set; respectively training a plurality of view classifiers on the marked data set based on different views; judging and marking partial data in the unmarked data set by using theplurality of trained view classifiers; adding the marked data into a marked data set to form a new marked data set; circularly carrying out the action of the new marked data set until a preset stop condition is reached; fusing the plurality of view classifiers to generate a secondary classifier; and judging and marking the data by using the secondary classifier. According to the method, the interaction, advantage complementation and collaborative learning among views are exerted by utilizing different learning difficulties of the data in different views, so that the effect of performing fraudmarking by adopting less and lower-cost sample data is achieved.

Random forests are one of the most classical machine learning algorithms and have been widely applied. However, observation finds that although numerous multi-view data exist in reality and wide analytical studies have been obtained, it is surprising that random forest construction for a multi-view scene is very few. Only methods for solving the multi-view learning problem by using random forestsare used for generating respective random forests for each view, and then multi-view information is fused during decision making. One significant defect of the method is that the correlation among multiple views is not utilized in the construction stage of the random forest, so that information resources are undoubtedly wasted. In order to make up the defect, the invention provides an improved multi-view learning algorithm based on random forest. Specifically, view fusion is carried out in the generation process of decision trees, information interaction between views is fused into the construction stage of the decision trees, and utilization of complementary information between the views in the whole random forest generation process is achieved. In addition, decision boundaries with discrimination properties are generated for the decision tree through discriminant analysis, so that the algorithm is more suitable for classification.

The invention discloses a multi-view learning method and system for rhesus monkey eye movement decision decoding, and belongs to the field of multi-view decoding in intrusive brain-computer interfaces. The multi-view learning method comprises the steps that an eye movement decision decoding model comprising a characteristic node extraction network, a reinforced node extraction network and a predicting network is established; eye movement decision direction data obtained after encoding of the local field potential and action potential of a medial frontal cortex auxiliary eye area in a rhesus monkey intrusive brain-computer interface, and one-hot encoding are input into the eye movement decision decoding model for training, and thus the trained eye movement decision decoding model is obtained; and the to-be-decoded local field potential and action potential are input into the trained decoding model, and thus a decoding result of the eye movement direction is obtained. The multi-view learning method and system are less in using limitation, low in computation complexity and suitable for decoding of action potential and local field potential signals in the intrusive brain-computer interface and other various multi-view learning scenes.

The invention discloses a color face recognition method based on semi-supervised multi-view dictionary learning. The method applies multi-view learning technology to dictionary learning of semi-supervised color face images. In the training phase, the method learns the structured dictionaries of each color component separately and makes these dictionaries orthogonal to each other to remove the correlation between each color component, making full use of the complementary color difference information between each color component; and the method In the dictionary learning process, color face image samples without class labels are used to participate in the training, and the information of all training samples is fully utilized. In the classification test stage, the method accumulates the reconstruction errors of each color component and uses the dictionary corresponding to each type of training samples to reconstruct the test samples, and finally classifies the test samples into the class with the smallest cumulative reconstruction error. The recognition effect of the invention is higher, and the color face recognition ability is obviously enhanced through the semi-supervised multi-view dictionary learning.

The invention discloses a short-term traffic forecasting method based on a multi-task multi-view learning model. The method comprises the following steps: 1, constructing a space-time data model individually for each road segment; 2, constructing a multi-core learning model; 3, constructing an objective function by adopting a multi-task multi-view feature learning model; 4, introducing a particleswarm optimization algorithm to optimize the objective function obtained in the step 3; and 5, repeating the steps 1 and 2 for any road segment to obtain input features, bringing the input features into an optimized objective function, and realizing short-term traffic prediction for any road segment. The method realizes the efficient prediction of short-term traffic, solves the problem that the spatial-temporal heterogeneity and the global prediction ability of the model can not reach equilibrium, solves the problem of parameter optimization of the model, and can be widely used in urban planning, human mobility survey, automobile navigation, emergency response, space-time accessibility analysis and traffic pollution modeling.

The present invention relates to the technical field of the Internet, and more specifically, relates to a webpage classification method for semi-supervised multi-view learning, comprising: obtaining data from webpages and establishing a training set; training a classifier through a marked training set; The machine codes the marked training set and the unmarked training set to obtain the sample features; performs density clustering on the sample features to obtain the clustering results; and classifies the samples of the unmarked training set according to the clustering results. This scheme uses the labeled training set to train the classifier, adds orthogonal constraints and confrontational similarity constraints on the basis of the existing multi-view classification method, and then uses the trained classifier to perform density aggregation on all the data in the training set. Finally, the accuracy of the classifier is verified, and the above-mentioned process can be iterated multiple times to improve the classification performance of the classifier.

The invention relates to a search engine user information demand satisfaction evaluation method that combines multi-view and semi-supervised learning. The user satisfaction model of supervised learning and six stages of evaluation, the present invention uses a small amount of labeled data and a large amount of unlabeled data to improve the performance of the evaluation model through the method of semi-supervised learning, and introduces the idea of ​​multi-view learning to overcome the traditional Semi-supervised learning methods of views are prone to fall into the problem of local optima. The beneficial effects are: (1) the user satisfaction of search engine information needs can be effectively evaluated with a small amount of labeled data; (2) the user satisfaction model evaluation can be improved by using a small amount of labeled data and a large amount of unlabeled data performance; (3) Describe the user's search process from the perspective of behavior and time, and avoid the model from falling into local optimum through mutual learning.

The invention discloses a method, device and medium for complementing missing data in time and space based on multi-view learning. The method aims at massive and multi-type environmental monitoring data sets, and constructs each data missing monitoring site including monitoring site, monitoring time, etc. , monitoring parameter data tensor in three dimensions, and decompose three mixed view matrices, use the matrix complete algorithm to estimate the missing data from the local and global scales for the three mixed views constructed, and then perform multiple View learning to obtain the estimation results of the missing values ​​of the final monitoring parameters. The invention can complete the missing value of the environmental monitoring data set, and has important significance in the fields of marine environment monitoring, smart city development and the like.