The nonnegative tensor （matrix） factorization finds more and more applications in various disciplines including machine learning, data mining, and blind source separation, etc. In computation, the optimization probl...The nonnegative tensor （matrix） factorization finds more and more applications in various disciplines including machine learning, data mining, and blind source separation, etc. In computation, the optimization problem involved is solved by alternatively minimizing one factor while the others are fixed. To solve the subproblem efficiently, we first exploit a variable regularization term which makes the subproblem far from ill-condition. Second, an augmented Lagrangian alternating direction method is employed to solve this convex and well-conditioned regularized subproblem, and two accelerating skills are also implemented. Some preliminary numerical experiments are performed to show the improvements of the new method.展开更多
Owing to its efficiency in solving some types of large-scale separable optimization problems with linear constraints, the convergence rate of the alternating direction method of multipliers(ADMM for short) has recentl...Owing to its efficiency in solving some types of large-scale separable optimization problems with linear constraints, the convergence rate of the alternating direction method of multipliers(ADMM for short) has recently attracted significant attention. In this paper, we consider the generalized ADMM(G-ADMM), which incorporates an acceleration factor and is more efficient. Instead of using a solution measure that depends on a bounded set and cannot be easily estimated, we propose using the original ?-optimal solution measure, under which we prove that the G-ADMM converges at a rate of O(1/t). The new bound depends on the penalty parameter and the distance between the initial point and the solution set, which is more reasonable than the previous bound.展开更多
数独是一个难以求解的整数规划问题,可以通过实数编码的方式去除整数约束的限制,将整数规划模型转化为一个ℓ_(0)范数极小化模型.已有算法大多是求解松弛的ℓ1范数极小化模型,只能求解部分数独问题.本文证明对于数独这样一个特殊的问题,ℓ_...数独是一个难以求解的整数规划问题,可以通过实数编码的方式去除整数约束的限制,将整数规划模型转化为一个ℓ_(0)范数极小化模型.已有算法大多是求解松弛的ℓ1范数极小化模型,只能求解部分数独问题.本文证明对于数独这样一个特殊的问题,ℓ_(q)(0<q<1)范数极小化模型等价于ℓ_(0)范数极小化模型,同时用ℓ_(1/2)-SLP(sequential linear programming)算法求解ℓ_(1/2)范数极小化模型.数值实验表明该方法可以求解更多的数独问题,本文从时间和成功率两方面验证了算法的高效性.展开更多
文摘The nonnegative tensor （matrix） factorization finds more and more applications in various disciplines including machine learning, data mining, and blind source separation, etc. In computation, the optimization problem involved is solved by alternatively minimizing one factor while the others are fixed. To solve the subproblem efficiently, we first exploit a variable regularization term which makes the subproblem far from ill-condition. Second, an augmented Lagrangian alternating direction method is employed to solve this convex and well-conditioned regularized subproblem, and two accelerating skills are also implemented. Some preliminary numerical experiments are performed to show the improvements of the new method.
基金supported by a Project Funded by the Priority Academic Program Development of Jiangsu Higher Education InstitutionsNational Natural Science Foundation of China (Grant Nos. 11401315, 11625105, 11171159 and 11431102)the National Science Foundation from Jiangsu Province (Grant No. BK20140914)
文摘Owing to its efficiency in solving some types of large-scale separable optimization problems with linear constraints, the convergence rate of the alternating direction method of multipliers(ADMM for short) has recently attracted significant attention. In this paper, we consider the generalized ADMM(G-ADMM), which incorporates an acceleration factor and is more efficient. Instead of using a solution measure that depends on a bounded set and cannot be easily estimated, we propose using the original ?-optimal solution measure, under which we prove that the G-ADMM converges at a rate of O(1/t). The new bound depends on the penalty parameter and the distance between the initial point and the solution set, which is more reasonable than the previous bound.
文摘数独是一个难以求解的整数规划问题,可以通过实数编码的方式去除整数约束的限制,将整数规划模型转化为一个ℓ_(0)范数极小化模型.已有算法大多是求解松弛的ℓ1范数极小化模型,只能求解部分数独问题.本文证明对于数独这样一个特殊的问题,ℓ_(q)(0<q<1)范数极小化模型等价于ℓ_(0)范数极小化模型,同时用ℓ_(1/2)-SLP(sequential linear programming)算法求解ℓ_(1/2)范数极小化模型.数值实验表明该方法可以求解更多的数独问题,本文从时间和成功率两方面验证了算法的高效性.
