44 results about "Subgradient method" patented technology

Computationally efficient accelerated D2-clustering algorithms are disclosed for clustering discrete distributions under the Wasserstein distance with improved scalability. Three first-order methods include subgradient descent method with re-parametrization, alternating direction method of multipliers (ADMM), and a modified version of Bregman ADMM. The effects of the hyper-parameters on robustness, convergence, and speed of optimization are thoroughly examined. A parallel algorithm for the modified Bregman ADMM method is tested in a multi-core environment with adequate scaling efficiency subject to hundreds of CPUs, demonstrating the effectiveness of AD2-clustering.

The optimal power allocation method in the multi-channel cognitive wireless network based on convex optimization disclosed by the present invention establishes a multi-channel cognitive wireless network model based on the frequency-common sharing mechanism, and establishes the total transmission power in peak form and each The optimization problem of maximizing the total communication rate of secondary users under the constraint of interference power of multiple channels, the convexity of the optimization problem is verified by using the convex optimization theory, and it is decomposed into sub-problems under specific channel conditions, and the Lagrange The dual method gives the KKT condition that the optimal solution satisfies, and gives the expression of the optimal solution with respect to the dual variable. Finally, the optimal power allocation scheme is solved by using the dichotomy method and the subgradient method. The present invention jointly considers multiple channels for spectrum sharing, adopts instantaneous peak interference power constraints to strictly protect primary users from interference, and uses multiple channels to enable secondary users to use greater transmission power to transmit, thereby further improving the comprehensiveness of spectrum resources utilization rate.

The invention discloses a high-resolution SAR image classification method based on a co-sparse model. The invention solves the technical problem that the SAR image classification is limited to expressing images with a comprehensive sparse model, resulting in high classification time complexity. The process is as follows: select the initial pixel value matrix X in the SAR image to be classified; select the analytic operator to learn the initial sample; combine the projection subgradient method and the unified row norm tight frame method to learn the analytic operator Ω; use the augmented Lager The Langer method is used to solve the co-sparse coefficient Z; the co-sparse coefficient vector of each pixel corresponding to the pixel block is combined with the pixel value vector of the pixel block to obtain the feature vector; based on the SVM classifier classification, the feature vector of each pixel in the whole image is obtained The predicted label; display the predicted label result with a grayscale image. The invention can quickly obtain the sparse representation of the image, ensures the timeliness and classification accuracy of the SAR image classification, and is used for the classification of the high-resolution SAR image.