32 results about "Group lasso" patented technology

A feature extraction and feature selection method oriented to background multi-source data is characterized by including the following steps that first, background data of a plurality of months are divided into a training set and a testing set; second, corresponding grouping features are extracted on the training set according to different source data; third, feature group selection is performed on the testing set through cross validation with a Group Lasso method. The feature extraction and feature selection method oriented to background multi-source data has the advantages that a C45 strategy tree is used for the selected group feature so that a classifier off-network user analysis classifier can be established, the prediction accuracy on off-network users reaches 45%, and the prediction accuracy on downtime users with the off-network tendency reaches 88%.

A method for constructing a dictionary to represent data from a training data set comprising: modeling the data as a linear combination of columns; modeling outliers in the data set via deterministic outlier vectors; formatting the training data set in matrix form for processing; defining an underlying structure in the data set; quantifying a similarity across the data; building a Laplacian matrix; using group-Lasso regularizers to succinctly represent the data; choosing scalar parameters for controlling the number of dictionary columns used to represent the data and the number of elements of the training data set identified as outliers; using BCD and PG methods on the vector-matrix-formatted data set to estimate a dictionary, corresponding expansion coefficients, and the outlier vectors; and using a length of the outlier vectors to identify outliers in the data.

The invention relates to the technical field of infrared spectrum wavelength, and concretely relates to a convex optimization theory-based novel infrared spectrum wavelength selection method. The method is a self-adaptive wavelength selection method adopting a Group Lasso technology without knowing spectrum partitioning priori knowledge. An infrared spectrum wavelength screening problem is converted into a Group Lasso sparse optimization problem, the sparse structure priori knowledge of the infrared spectrum is fully used to self-adaptively determine the partitioning size, a Shooting rapid algorithm is adopted to calculate the sparse solution, and Belsley colinearity examination of the sparse solution is carried out on the sparse solution to reject wavelength points with small contribution. The method has the advantages of small calculation capacity, few adjustable parameters and strong robustness, can effectively reduce the complexity of a model and improve the generalization performance of the model, and can be widely used in the field of infrared spectrum wavelength selection of a solid phase, a liquid phase and a gas phase.

The invention discloses a matrix and tensor combined decomposition-based car recommendation method and system. The method comprises the following steps of constructing a car scoring tensor, constructing a car manufacturer and supplier relationship matrix, and constructing a car product structure tree, wherein the car manufacturer and supplier relationship matrix and the car product structure are both complete and used for assisting in predicting a missing specific value in the tensor; according to the car product structure tree, introducing a tree-guided group lasso model for performing stipulation on a final loss function to obtain a weight; establishing a loss function, and performing iteration on the loss function by using an alternating least square method; performing derivation and zero setting on the loss function, and then calculating an iterative function of the matrix; recovering the tensor, namely complementing a missing value of the tensor; and for different users, based on elements in the complemented tensor X, recommending favorite car types of the users to the users according to a sequence of scores from high to low.

The invention provides a semi-supervised hash image search method based on Group Lasso, and belongs to the field of image search. The method comprises the steps of identifying label images and non-label images in an image database, and pre-processing input images, the label images and the non-label images; according to the pre-processed input images, label images and non-label images, conducting semi-supervised hash learning based on the Group Lasso to obtain binary hash code corresponding to the images; according to the binary hash codes, calculating the Hamming distances between the input images and the images in the image database, and returning the image corresponding to the minimum Hamming distance as an image research result. In the semi-supervised hash image search method based on the Group Lasso, in combination with the condition of current image data, an image data structure can be effectively modeled, a desired image can be quickly and accurately found, and the storage space is greatly saved without storing the image.

The invention discloses a multi-fault-feature identification method based on a sparse multiperiod-group lasso. The multi-fault-feature identification method comprises the following steps that an to-be-identified signal is analyzed so as to construct a binary periodic sequence b, based on a fault feature signal, the between-group sparse characteristic in period groups are presented to obtain a regularization term P (x;b) for promoting between-group sparseness in the period groups, and a sparse multiperiod-group lasso model is established based on discrimination of different fault feature frequencies; controlled optimization operators of a data fidelity term (please see the specifications for the formula) and the regularization term (please see the specifications for the formula) in the sparse multiperiod-group lasso model are constructed correspondingly, through decoupling of the controlled optimization operators, variables are separated, aiming at each controlled optimization operator,the closed-form solution optimized by the controlled optimization operator is established, through iteration, the closed-form solution corresponding to the controlled optimization operator of each fault is solved, and thus model solving is achieved; regularization parameters are set adaptively through simulation signal counting and analyzing, the adaptive solution of the algorithm is obtained through the parameters, and thus each fault is obtained through separation; and aiming at each fault obtained through separation, the fault type is identified through envelope analysis.

The invention belongs to the field of biological information, and discloses a cancer classification and characteristic gene selection method, which comprises the following steps of: establishment of a primary learner: establishing T logistic regression models and a spark group lasso regularized loss function solving model corresponding to the T logistic regression models, and outputting a secondary learner training set; establishing a secondary learner: establishing a multi-response regression model and a loss function solving model corresponding to L1 regularization, and outputting a training set prediction result; and a prognosis feature selection model: establishing a prognosis feature selection SGL model. According to the cancer classification and feature gene selection method, the three standards of prediction, stabilization and selection are met, the accuracy and stability of the model on cancer classification prediction are improved through stacking integration, oncogenes and cancer-related genes are accurately selected, and the interpretability of the model is enhanced; gene and gene pathway priori knowledge are fused, and the accuracy of cancer classification and the effectiveness of feature selection are improved.

Described herein are systems and methods to prune deep neural network models in reducing the overall memory and compute requirements of these models. It is demonstrated that using block pruning and group lasso combined with pruning during training, block-sparse recurrent neural networks (RNNs) may be built as accurate as dense baseline models. Two different approaches are disclosed to induce blocksparsity in neural network models: pruning blocks of weights in a layer and using group lasso regularization to create blocks of weights with zeros. Using these techniques, it is demonstrated that block-sparse RNNs with high sparsity can be created with small loss in accuracy. Block-sparse RNNs eliminate overheads related to data storage and irregular memory accesses while increasing hardware efficiency compared to unstructured sparsity.

The invention discloses a small sample-based genotype and phenotype correlation analysis method in multi-omics data. The method specifically comprises the following steps: generating a weighted undirected gene association graph by using a protein network and a gene expression value, and clustering the undirected graph by using an SPICi clustering method to generate a gene cluster; screening the gene clusters by using a group Lasso method; obtaining an SNP cluster corresponding to the screened gene cluster through the eQTL data; constructing each SNP cluster, the corresponding gene cluster and the phenotype into a three-layer network class block, performing regression operation on the association relationship between the SNP and the gene in each class block by adopting a sparse partial least square method, and performing operation on the association relationship between the gene and the phenotype by adopting logistic regression; and averaging the obtained prediction results of the blocks to obtain a final prediction result. The method can solve the problem that effective regression cannot be realized due to huge characteristic values under the condition of small samples in a three-layer network; wherein the prediction accuracy is improved, the biological significance is clearer and tissue specificity is considered.

The invention relates to an enhancement strategy of a high-order factor decomposer model, in particular to a high-order factor decomposer method (simply written as SGL-HOFM) based on sparse group Lasso. The high-order factor decomposer method is characterized in that a Sparse Group Lasso-based regular item is adopted for a special intersection item structure of a high-order factor decomposition machine instead of a traditional two-norm, and a new model can realize sparse structure of features and play a role in feature selection. After the high-order factor decomposer method is adopted, the number of parameters of the model is obviously reduced, so that the required storage space is reduced; due to the fact that noise is removed through feature selection, the performance of the model is improved; and the high-order factor decomposer method can be widely applied to the field of data mining, such as classification, regression, sorting, feature analysis, a recommendation system and socialnetwork analysis, and is particularly suitable for solving the analysis problem containing large-scale sparse high-dimensional data.