92 results about "Coordinate descent" patented technology

A method recovers an uncorrupted low-rank matrix, noise in corrupted data and a subspace from the data in a form of a high-dimensional matrix. An objective function minimizes the noise to solve for the low-rank matrix and the subspace without estimating the rank of the low-rank matrix. The method uses group sparsity and the subspace is orthogonal. Random subsampling of the data can recover subspace bases and their coefficients from a much smaller matrix to improve performance. Convergence efficiency can also be improved by applying an augmented Lagrange multiplier, and an alternating stepwise coordinate descent. The Lagrange function is solved by an alternating direction method.

A method and system for image reconstruction of data acquired by a device such as computed tomography is provided. The method and system use a multi-stage statistical iterative reconstruction techniques to provide a three dimensional representation of the scanned object. In one embodiment, the first stage uses a projection-based reconstruction technique, such as Ordered Subset (OS) to converge on a solution for low frequency portion of the image. A subsequent stage uses a voxel-based reconstruction technique, such as Iterative Coordinate Descent (ICD), to converge on a solution for high frequency portions of the image. Systems and methods for reconstructing images from incomplete or partial projection data is also provided.

Optimal enhancement of 3G cellular network capacity utilizes two components of learning and optimization. First, a pair of learning approaches are used to model cellular network capacity measured in terms of total number of users carried and predict breakpoints of cellular towers as a function of network traffic loading. Then, an optimization problem is formulated to maximize network capacity subject to constraints of user quality and predicted breakpoints. Among a number of alternatives, a variant of simulated annealing referred to as Block Coordinated Descent Simulated Annealing (BCDSA) is presented to solve the problem. Performance measurements show that BCDSA algorithm offers dramatically improved algorithmic success rate and the best characteristics in utility, runtime, and confidence range measures compared to other solution alternatives. Accordingly, integrated iterative method, program, and system are described aiming at maximizing the capacity of 3G cellular networks by redistributing traffic from congested cellular towers to non-congested cellular towers.

The invention discloses a maximal posteriori optimized image reconstruction method for leading in a general Gibbs experiment in the PET imaging. The method comprises the following procedures: (1) PET imaging equipment is utilized to collect detector data before imaging, and the corrective parameter value and the system matrix of various data in the imaging equipment are obtained simultaneously; (2) a mathematical statistic statistical model used for reconstructing an image is reconstructed according to a statistical feature that is met by the corrective data acquainted by the procedure (1) before imaging; (3) the general Gibbs experiment is led in aiming at the compute of a mathematical model in the procedure (2) , a maximal posteriori estimate method is adopted to perform the conversion of a reconstruction model, to obtain an optimized equation with a constrained objective function used for obtaining a PET reconstruction image; (4) a parabola is adopted to replace a coordinate descent algorithm, to perform the iterative computation treatment and to reconstruct the image based on the selection of a global parameter in the optimized equation through a result obtained by the procedure (3). The invention can greatly improve the quality of the PET reconstruction image.

The invention discloses a synthetic aperture radar self-focusing method. The synthetic aperture radar self-focusing method specifically comprises the following steps: range pulse compression, range migration correction, azimuth de-ramping, coordinate descent multidimensional phase error estimation, single azimuth unit phase error estimation and multidimensional phase error estimation value output. The synthetic aperture radar self-focusing method disclosed by the invention has the advantages that since a coordinate descent method is used for conducting multidimensional joint estimation of phase errors of each azimuth at each moment and a multidimensional projection method is used for transforming the problem of solving maximum contrast by one-dimension search into the problem of solving a quartic polynomial, the calculation amount is greatly reduced; the estimation of azimuth phase errors by adopting the method disclosed by the invention has the feature of no order limit.

The invention provides a method for distribution of time-frequency resources of unmanned aerial vehicles based on block coordinate descent. The method comprises the following steps of determining thenumber of the unmanned aerial vehicles, setting a system authorization channel set, and setting a maximum power upper limit of the system and other environmental parameters; constructing a spectrum allocation matrix and an adjacent channel interference matrix to be introduced into an original problem, and utilizing the structural characteristics of the problem to simplify an original complex max-min hybrid planning problem; providing a low-complexity iterative algorithm based on the block coordinate descent to solve the problem; and making a ground control station carry out power and channel distribution on the unmanned aerial vehicles through a remote control channel according to a maximum SINR value of each unmanned aerial vehicle, and completing the design of the method for distributionof the time-frequency resources of the unmanned aerial vehicles. The method for distribution of the time-frequency resources of the unmanned aerial vehicles based on the block coordinate descent hasthe beneficial effects that the reasonable distribution of the limited frequency spectrum resources is realized on the premise of improving the reliability of a control signal received by each unmanned aerial vehicle and ensuring the maximum consumption power constraint of the system.

The invention discloses an SINR maximization method for an IRS-assisted dual-function radar communication system, and the method achieves the maximization of the SINR of an MIMO radar through the joint transmission beam forming and phase shift optimization technology under the condition of considering the existence of an eavesdropping target in an intelligent reflection surface (IRS)-assisted multiple-input-multiple-output (MIMO) radar. According to the method, an IRS-assisted dual-function radar communication (IDFRC) system considering a physical layer security (PLS) technology is established, a transmitting beam former, an artificial noise vector and the phase shift of an IRS are jointly optimized to maximize the SINR of an MIMO radar, and in order to solve the optimization problem non-convexity caused by the non-concavity of a security rate function, a block coordinate descent (BCD) algorithm is used to alternately update the transmit power and the phase shift of the IRS. Specifically, a controlled minimization (MM) algorithm is utilized to optimize a phase shift at a given transmit power, a first-order Taylor expansion is utilized to convert a problem into a convex optimization problem, and two transmit beamforming vectors are designed to detect a target and securely transmit information to a legal receiver. The method has the advantages that the complexity is low, the signal to interference plus noise ratio of the radar is remarkably improved, and the feasibility and the practicability are high.

The invention discloses an unmanned aerial vehicle auxiliary edge calculation unloading method based on terminal energy efficiency optimization, and the method comprises the steps: jointly optimizingthe task load of local calculation of a user terminal, the task load of calculation unloading and the track of an unmanned aerial vehicle through a global optimization algorithm based on block coordinate descent; the optimization effect on the energy consumption of the user terminal under different channel conditions is discussed, and finally, the unmanned aerial vehicle auxiliary edge calculationmodel taking the minimization of the energy consumption of the user terminal as a target and taking the task load and the time delay demand of the user and the energy of the unmanned aerial vehicle as constraints is constructed. According to the method, computing resources of a local side and an edge side are fully utilized, the energy efficiency of the user terminal is improved, the time delay of local calculation of the user terminal is considered, synchronous completion of local and edge side calculation in task time is ensured, and the effectiveness of the method is improved. In addition,effectiveness and reliability of the method under different channel conditions are verified, and the method has a wider practical application scene.

The invention discloses a distributed optimal power flow method based on continuous punishment dual decomposition for a power system. The distributed optimal power flow method comprises the following steps: firstly acquiring network parameters of the power system, and determining a mathematical model of the optimal power flow problem; secondly introducing an auxiliary variable to obtain an equivalent problem of the original optimal power flow control problem; then solving the equivalent problem through inner and outer iterative algorithms, wherein the inner iteration is used for carrying out distributed solution on a corresponding inner augmented lagrangian problem by using a block coordinate descent algorithm under the condition of a fixed dual variable, and each bus needs to conduct data interaction with a neighboring bus before the optimization of each group of variables to realize local optimization of a local variable, and the outer iteration is used for updating a dual variable and a punishment factor according to the current constraint feasibility index; finally, completing the optimal power flow control over the power system according to an obtained bus input power value. According to the distributed optimal power flow method based on the continuous punishment dual decomposition for the power system, the optimal power flow of the power system is subjected to distributed design by using a continuous punishment dual decomposition technology, so that the performance loss of the system is the least on the premise of guaranteeing the feasibility of a power flow equation.

Time-of-flight (TOF) clinical data collected during a PET scan are very sparse and have significant size. These data undergo TOF axial rebinning and azimuthal mashing if histogrammed data-based reconstruction algorithms are used. In a clinical environment, TOF compression is typically performed by the hardware rebinner. Normalization data, acquired on a regular basis and used for estimation of some norm components, are compressed by the hardware rebinner in a similar manner. This disclosure presents simple update iterative algorithms for crystal efficiencies norm component estimation from TOF compressed normalization data. Previously known methods are not directly applicable since the compression procedure significantly complicates normalization data model equations. The iterative algorithms presented herein have advantages of being easily adapted to any acquisition geometry, and of allowing estimation of parameters at crystal level when a number of crystals is relatively small. A monotonic sequential coordinate descent algorithm, which optimizes the Least Squares objective function, is presented. A simultaneous update algorithm, which possesses the advantage of easy parallelization, is also presented.

The invention discloses a computing efficiency optimization method in a millimeter wave mobile edge computing system based on a reconfigurable intelligent surface, and the method comprises the steps: building a computing efficiency optimization model based on a maximum-minimum fairness criterion in a partial unloading mode; performing combined optimization on hybrid beam forming of a base station, passive beam forming of a reconfigurable intelligent surface, transmitting power of each user and frequency of a local central processing unit, and giving a calculation efficiency optimization algorithm based on penalty non-precision block coordinate descent to obtain the combined optimization method. The joint optimization method provided by the invention is effective.

The present disclosure provides a new scalable coordinate descent (SCD) algorithm and associated system for generalized linear models whose convergence behavior is always the same, regardless of how much SCD is scaled out and regardless of the computing environment. This makes SCD highly robust and enables it to scale to massive datasets on low-cost commodity servers. According to one aspect, by using a natural partitioning of parameters into blocks, updates can be performed in parallel a block at a time without compromising convergence. Experimental results on a real advertising dataset are used to demonstrate SCD's cost effectiveness and scalability.

Optimal enhancement of 3G cellular network capacity utilizes two components of learning and optimization. First, a pair of learning approaches are used to model cellular network capacity measured in terms of total number of users carried and predict breakpoints of cellular towers as a function of network traffic loading. Then, an optimization problem is formulated to maximize network capacity subject to constraints of user quality and predicted breakpoints. Among a number of alternatives, a variant of simulated annealing referred to as Block Coordinated Descent Simulated Annealing (BCDSA) is presented to solve the problem. Performance measurements show that BCDSA algorithm offers dramatically improved algorithmic success rate and the best characteristics in utility, runtime, and confidence range measures compared to other solution alternatives. Accordingly, integrated iterative method, program, and system are described aiming at maximizing the capacity of 3G cellular networks by redistributing traffic from congested cellular towers to non-congested cellular towers.

The invention discloses a Rayleigh wave inversion method of employing simulated annealing improved on the basis of differential evolution and block coordinate descent. The method comprises the steps of firstly building an inversion objective function; optimizing a parameter model of the inversion objective function by using an differential evolution algorithm and obtaining best individuals; carrying out simulated annealing treatment on each individual by adopting a block coordinate descent algorithm; and taking the obtained new individuals as the individuals of the next generation population.According to the Rayleigh wave inversion method, the accuracy of simulated annealing inversion can be effectively improved and the controllability of simulated annealing on the accuracy is achieved through building the inversion objective function, optimizing the inversion objective function by adopting the differential evolution algorithm and dividing a multi-parameter problem into a local iterative optimization problem of multiple single parameters by adopting the block coordinate descent algorithm, thereby improving the stability of simulated annealing; and the Rayleigh wave inversion method can be applied to inversion of seismic data of actual shallow exploration, and the prediction accuracy of a near-surface underground structure is improved.

A method recovers an uncorrupted low-rank matrix, noise in corrupted data and a subspace from the data in a form of a high-dimensional matrix. An objective function minimizes the noise to solve for the low-rank matrix and the subspace without estimating the rank of the low-rank matrix. The method uses group sparsity and the subspace is orthogonal. Random subsampling of the data can recover subspace bases and their coefficients from a much smaller matrix to improve performance. Convergence efficiency can also be improved by applying an augmented Lagrange multiplier, and an alternating stepwise coordinate descent. The Lagrange function is solved by an alternating direction method.

The invention discloses a multi-level optimized grid curved surface discrete spline curve design method. According to the method, the constraint that a curve is strictly located on a curved surface isrelaxed, only discrete sampling points of the curve are located on a grid curved surface, the thought of an interior point method is adopted, and numerical solution is conducted through a global optimization method based on a block coordinate descent method. In order to more accurately estimate a discrete differential operator, improve the solving precision, reduce the calculated amount of movingsampling points and improve the solving efficiency, a coarse-to-fine multi-scale hierarchical solving strategy is adopted, that is, the density of curve sampling points is gradually increased for iterative solving until the algorithm converges. And finally, mapping the curve segments to the grid curved surface by means of local parameterization. A convergence analysis experiment shows that the multi-level optimization method can achieve rapid convergence, and a smoother result is obtained under a multi-scale strategy. Compared with an existing projection method and an existing fairing method,the method is higher in efficiency and has certain advantages in controllability, universality and robustness.