98 results about "Learning set" patented technology

In a pre-processing step prior to training a learning machine, pre-processing includes reducing the quantity of features to be processed using feature selection methods selected from the group consisting of recursive feature elimination (RFE), minimizing the number of non-zero parameters of the system (l_o-norm minimization), evaluation of cost function to identify a subset of features that are compatible with constraints imposed by the learning set, unbalanced correlation score and transductive feature selection. The features remaining after feature selection are then used to train a learning machine for purposes of pattern classification, regression, clustering and/or novelty detection.

Search results are processed using search requests, including analyzing received queries in order to provide a more sophisticated understanding of the information being sought. A concept network is generated from a set of queries by parsing the queries into units and defining various relationships between the units. From these concept networks, queries can be automatically categorized into categories, or more generally, can be associated with one or more nodes of a taxonomy. The categorization can be used to alter the search results or the presentation of the results to the user. As an example of alterations of search results or presentation, the presentation might include a list of “suggestions” for related search query terms. As other examples, the corpus searched might vary depending on the category or the ordering or selection of the results to present to the user might vary depending on the category. Categorization might be done using a learned set of query-node pairs where a pair maps a particular query to a particular node in the taxonomy. The learned set might be initialized from a manual indication of which queries go with which nodes and enhanced has more searches are performed. One method of enhancement involves tracking post-query click activity to identify how a category estimate of a query might have varied from an actual category for the query as evidenced by the category of the post-query click activity, e.g., a particular hits of the search results that the user selected following the query. Another method involved determining relationships between units in the form of clusters and using clustering to modify the query-node pairs.

An intelligent instrumented air cleaner comprises a multi-functional sensor, a central processor, a cloud server, a wireless communication module, a purification device, an alarm device, a motor driving mechanism and a humidifier. According to the intelligent instrumented air cleaner, updating and exchange of memory data are realized through the connection between the wireless communication module and a cloud server of the Internet of Things, the air cleaner is controlled through the connection between a mobile device and the cloud server so that a remote controller can be replaced, remote network control is also realized, the cloud server also has an analysis function and a processing result advice providing function and sends processed data to users for reference, the users can set the working state of the air cleaner by themselves and define working strategies by themselves, the air cleaner saves the strategies automatically and performs actions according to the strategies, and self-selection of self-learning setting optimization is conducted on the data in a memory through big data analysis and processing. By the adoption of the intelligent instrumented air cleaner, an intelligent instrumented household net and an intelligent instrumented office net are realized in the true sense, air is purified, safe and reliable living and working environments are provided, useless consumption of electric energy of a traditional air cleaner is avoided, energy consumption is reduced, using pleasure of a user is improved, the parameters of the environment where the user is located are provided for the user in time, and reliable parameters and solutions are provided for environment change handling.

In a pre-processing step prior to training a learning machine, pre-processing includes reducing the quantity of features to be processed using feature selection methods selected from the group consisting of recursive feature elimination (RFE), minimizing the number of non-zero parameters of the system (l₀-norm minimization), evaluation of cost function to identify a subset of features that are compatible with constraints imposed by the learning set, unbalanced correlation score and transductive feature selection. The features remaining after feature selection are then used to train a learning machine for purposes of pattern classification, regression, clustering and/or novelty detection.

The invention discloses an advertisement click-through rate prediction method based on multi-dimensional feature combination logical regression. The method comprises the first step that feature information of a hierarchical structure of the user hierarchy, feature information of a hierarchical structure of the media hierarchy and feature information of a hierarchical structure of the advertisement hierarchy are extracted from the obtained click-through rate data respectively; the second step that multi-dimensional combination is carried out on the feature information of the hierarchical structure of the user hierarchy, the feature information of the hierarchical structure of the media hierarchy and the feature information of the hierarchical structure of the advertisement hierarchy, three-to-three combination is carried out on one-dimensional feature information in the feature information to obtain a three-dimensional feature combination, and a feature vector combined by the three-dimensional feature information is formed to represent a user cluster; the third step that the second step is carried out repeatedly and a learning set of the feature vector combined by the three-dimensional feature information is obtained; the fourth step that the learning set obtained in the third step is used for training and testing a logical regression model, and the logical regression model is used for predicting the advertisement click-through rate.

The invention discloses a method for forecasting a ship wake based on an optimized support vector regression parameter, and the method comprises the following steps of: providing a training set firstly; selecting a type of a kernel function, and implementing a model training through a learning set to obtain a forecasting function; reading and judging whether a mean square error of the forecasting function reaches a precision requirement; if so, forecasting the ship wake through the function, if not, generating a parameter path searching map; searching an optimum parameter through an intelligent water drop algorithm until reaching the precision requirement. While searching an optimal parameter, the intelligent water drop has a positive feedback mechanism in an optimum searching process, therefore, a searching process is rapid and effective; a sand content of a path can change in accordance with the sand content of an optimum water drop after searching each generation of water drops, thereby avoiding a phenomenon of early maturing and improving a searching ability; a final forecasting function can be rapidly obtained; and rapid and effective forecasting of the ship wake is realized.

The invention discloses a network optimal selection method based on deep learning. The method comprises the following steps: performing a labeling operation on an acquired network link feature value so as to acquire a supervised learning data set; inputting the acquired supervised learning data set into a deep learning model to train so as to obtain a trained model; inputting a to-be-tested link data into the trained model to obtain an optimal path through an optimal path selection algorithm. By use of the optimal path selection method disclosed by the invention, the deep learning theory is innovatively and successfully applied to the practice of the network optimal path selection, and a good presentation result is obtained, an entire running efficiency of the network can be improved, and a certain push effect is provided for the future intelligent network development.

The invention discloses a coke furnace temperature predicting method based on dynamic working conditions in a coke furnace heating and burning process. The method is characterized by comprising the following steps: 1, acquiring gas flow and coke furnace temperature data in various historical working conditions, and establishing a sample bank; 2, calculating the similarity of a current working point and a sample in the sample bank on the basis of an included angle of Euclidean distance and variation trend of the current working point and the sample in the sample bank; and 3, selecting a plurality of sample structural learning sets with maximum similarity; establishing a learning-set-based local linear model by adopting an iterative least square method; and calculating an output value corresponding to the current working point as a coke furnace temperature predicted value in the coke furnace heating and burning process. The coke furnace temperature predicting method based on dynamic working conditions in the coke furnace heating and burning process has high prediction accuracy, and has the online adaptive capability.

The invention relates to the technical field of recommendation systems based on mass data and data mining, in particular to a personalized recommendation system and method. The system comprises a data interface layer, a user log system, a knowledge base, an entity relation gallery and a recommendation calculation system. The data interface layer is used for being in communication with an upper layer service system. The user log system includes all operation records of a user in an application system. The knowledge base is a set of all data in the application system and a learning set of the recommendation system. The entity relation gallery is used for storing the incidence relation between the user, data entities, properties and the like. The recommendation calculation system automatically recommends topic data which the user is interested in to the user by integrating the preference of the user and the weight of the user and according to a specific algorithm. By means of the personalized recommendation system and method, the problem of the cold start of the recommendation system and the problem that when interest of the user changes ceaselessly, the recommendation calculation complexity is increased are solved; the personalized recommendation system and method can be used for processing mass data.

Embodiments for generating appropriate data sets for learning to identify user actions. A user uses one or more applications over a suitable period of time. As the user uses the applications, a monitoring device, acting as a “man-in-the-middle,” intermediates the exchange of encrypted communication between the applications and the servers that serve the applications. The monitoring device obtains, for each action performed by the user, two corresponding (bidirectional) flows of communication: an encrypted flow, and an unencrypted flow. Since the unencrypted flow indicates the type of action that was performed by the user, the correspondence between the encrypted flow and the unencrypted flow may be used to automatically label the encrypted flow, without decrypting the encrypted flow. Features of the encrypted communication may then be stored in association with the label to automatically generate appropriately-sized learning set for each application of interest.

The invention discloses a word frequency based skip language model training method, relates to the technical field of machine translation and aims at solving the OOV problem of a statistical language model caused by linguistic data shortage in the prior art. The word frequency based skip language model training method comprises the steps that Chinese sentences are collected; the Chinese sentences are segmented; a learning set corpus is generated; statistics is conducted on vocabulary and word frequency in the learning set corpus to generate a Chinese vocabulary wt; statistics is conducted on phrases and the emerging times of phrases in the corpus to generate a 1-n Chinese phrase table pt0; a selective skip standard k is set, and k judgment is performed according to the statistical results of the word frequency in the Chinese vocabulary wt, and when the sum of the number of all the vocabulary with the emerging times k not greater than i accounts for above 60% of the number of all the vocabulary, k = i; linguistic model training is performed according to a Chinese sentence table pt2 to obtain a skip-ngram linguistic model. The word frequency based skip language model training method is used for obtaining a linguistic model probability table.

The invention discloses an electrocardiosignal processing method. The electrocardiosignal processing method includes the following steps that A, an original electrocardiosignal set theta is collected and preprocessed to construct an input electrocardiosignal signal theta'; B, a sample F' is extracted, a hidden layer is trained, and a self-learning set phi is output; C, a label vector G of the self-learning set phi is detected, and a node is output; D, an input layer, the hidden layer and an output layer of an artificial neural network are stacked, so that a complete electrocardiosignal disease monitoring artificial neural network is established. Thus, basic waveform positioning is not needed, dependence on artificially selected characteristics is avoided, the requirement for priori knowledge is low, important characteristics in data can be automatically and abstractly input, and the dimensionality of an input electrocardiosignal sample is reduced through PCA whitening dimensionality reduction applied during preprocessing.

The invention discloses a near-surface air temperature inversion method comprising the following steps: establishing an original data record set of an unmanned weather station; constructing a first sub-pattern learning set and a first sub-pattern validation set; and acquiring a second sub-pattern to a fth sub-pattern, performing near-surface air temperature inversion to acquire a near-surface air temperature inversion image map of a target zone, and performing error correction to acquire a corrected near-surface air temperature inversion image map. According to the near-surface air temperature inversion method disclosed by the invention, the near-surface air temperature inversion is performed by collecting actually-measured air temperature of the unmanned weather station, collecting meteorological satellite data, DEM data and astronomy and calendar rules and also adopting a super nonlinear algorithm, and the near-surface air temperature inversion image map is then calculated by using a high-performance computer. Results show that the near-surface air temperature inversion method disclosed by the invention is relatively high in pattern accuracy, high in result reliability and strong in generalization ability, and ensures that the interferences of clouds, terrains and the like can be overcome; and a constructed CPU+GPU heterogeneously-cooperative parallel computer ensures that the computation speed can be increased by more than 1000 times, so that the near-surface air temperature inversion method is convenient for large-area application and computing capacity expansion.

The invention discloses a method for segmenting a radiotherapy image by combining a deep neural network and a probability graph model. The method comprises the following steps: before image segmentation, forming a learning set by using a group of two-dimensional medical images, pre-segmented images and benchmark annotation data corresponding to the medical images so as to train a deep neural network model; during image segmentation, using a deep neural network model for processing an input image and a pre-segmented image to obtain a group of multi-channel score maps, each pixel channel uniquely corresponding to a certain region of interest, and a score value in the pixel channel representing the probability that a pixel belongs to the region of interest corresponding to the channel; usinga probability graph model method for each multi-channel score graph to generate a multi-channel annotation graph, wherein each channel image represents the position annotation of the region of interest corresponding to the channel; and for each original two-dimensional medical image, fusing the multi-channel annotation images into a final annotation image as an image segmentation result.

Automatically classifying security vulnerabilities in computer software applications by identifying candidate security vulnerabilities in a learning set including at least a first computer software application, classifying each of the candidate security vulnerabilities using predefined classifications, determining, for each of the candidate security vulnerabilities, values for predefined properties, creating a set of correlations between the property values and the classifications of the candidate security vulnerabilities, identifying a candidate security vulnerability in a second computer software application, determining, for the candidate security vulnerability in the second computer software application, values for the predefined properties, and using the set of correlations to classify the candidate security vulnerability in the second computer software application with a classification from the predefined classifications that best correlates with the property values of the candidate security vulnerability in the second computer software application.

The invention discloses a full-reference and no-reference image quality evaluation method with a unified structure, and relates to image processing. The method includes; Carrying out image segmentation and filtering decomposition processing; And designing a feature extraction network and a regression network. The distortion image quality can be evaluated, and the method can be applied to full-reference and no-reference image quality evaluation tasks. A convolutional neural network is adopted to learn the mapping from the image features to the main pipe quality score, thereby realizing an imagequality evaluation task. The method is established on the basis of deep learning machine learning set statistics, and has the characteristics that the structure of different image quality evaluationtasks is unchanged, the practicability and reliability are high, the portability is high, the subjective quality score conforming to human eye aesthetic appreciation can be predicted, and the labor cost of human eye evaluation is saved.

A method of training a natural language processing unit applies a candidate learning set to at least one component of the natural language unit. The natural language unit is then used to generate a meaning set from a first corpus. A second meaning set is generated from a second corpus using a second natural language unit and the two meaning sets are compared to each other to form a score for the candidate learning set. This score is used to determine whether to modify the natural language unit based on the candidate learning set.

The invention discloses a wavelet transform-based fine-granularity self-learning integration prediction method which comprises the following steps: by adopting time sequence decomposition based on wavelet transform, predicting time sequences of different variable coefficients with different granularities, so as to relatively precisely reveal the variation rules of the time sequences; with the combination of the time sequence decomposition, extracting characteristics of a plurality of related factors, sufficiently capturing main influence factors, and predicting future trend through rule statistics, thereby being rapid, convenient and simple; applying a model-based aggregation algorithm frame to the regression process, thereby enabling the model to have robustness which is relatively good when being compared with that of a single based learner; predicting with the combination of composite models based on wavelet transform, SVR and Ensemble, thereby obtaining prediction performance which is relatively precise when being compared with that of a conventional single model. The wavelet transform-based fine-granularity self-learning integration prediction method can be widely applied to the technical field of big-data mining and machine learning.