Joint chance-constrained optimization problems under discrete distributions arise frequently in financial management and business operations. These problems can be reformulated as mixed-integer programs. The size of reformulated integer programs is usually very large even though the original problem is of medium size. This paper studies an augmented Lagrangian decomposition method for finding high-quality feasible solutions of complex optimization problems, including nonconvex chance-constrained problems. Different from the current augmented Lagrangian approaches, the proposed method allows randomness to appear in both the left-hand-side matrix and the right-hand-side vector of the chance constraint. In addition, the proposed method only requires solving a convex subproblem and a 0-1 knapsack subproblem at each iteration. Based on the special structure of the chance constraint, the 0-1 knapsack problem can be computed in quasi-linear time, which keeps the computation for discrete optimization subproblems at a relatively low level. The convergence of the method to a first-order stationary point is established under certain mild conditions. Numerical results are presented in comparison with a set of existing methods in the literature for various real-world models. It is observed that the proposed method compares favorably in terms of the quality of the best feasible solution obtained within a certain time for large-size problems, particularly when the objective function of the problem is nonconvex or the left-hand-side matrix of the constraints is random.

We consider the problem of minimizing the sum of an average of a large number of smooth convex component functions and a possibly nonsmooth convex function that admits a simple proximal mapping. This class of problems arises frequently in machine learning, known as regularized empirical risk minimization (ERM). In this article, we propose mSRGTR-BB, a minibatch proximal stochastic recursive gradient algorithm, which employs a trust-region-like scheme to select stepsizes that are automatically computed by the Barzilai-Borwein method. We prove that mSRGTR-BB converges linearly in expectation for strongly and nonstrongly convex objective functions. With proper parameters, mSRGTR-BB enjoys a faster convergence rate than the state-of-the-art minibatch proximal variant of the semistochastic gradient method (mS2GD). Numerical experiments on standard data sets show that the performance of mSRGTR-BB is comparable to and sometimes even better than mS2GD with best-tuned stepsizes and is superior to some modern proximal stochastic gradient methods.

This letter is concerned with a robust minimum variance beamformer design. To hedge the mismatch between the true and the assumed steering vectors, a distributionally robust beamformer (DR-beamformer) is proposed. The tractable reformulation of this beamformer is developed. Compared with the existing robust beamformers (e.g., worst-case robust beamformer and Gaussian robust beamformer), the proposed robust beamformer does not assume full knowledge of the channel mismatch. Therefore, it is more flexible in practice and more general in formulation. In addition, the relationships of the proposed robust beamformer to the existing ones are investigated. The performance gain of the DR-beamformer over the other robust beamformers is highlighted through numerical simulations.

A receiver design problem for multi-access space-time block coded multiple-input multiple-output systems is considered. To hedge the mismatch between the true and the estimated channel state information (CSI), several robust receivers have been developed in the past decades. Among these receivers, the Gaussian robust receiver has been shown to be superior in performance. This receiver is designed based on the assumption that the CSI mismatch has Gaussian distribution. However, in real-world applications, the assumption of Guassianity might not hold. Motivated by this fact, a more general distributionally robust receiver is proposed in this paper, where only the mean and the variance of the CSI mismatch distribution are required in the receiver design. A tractable semi-definite programming (SDP) reformulation of the robust receiver design is developed. To suppress the self-interferences, a more advanced distributionally robust receiver is proposed. A tight convex approximation is given and the corresponding tractable SDP reformulation is developed. Moreover, for the sake of easy implementation, we present a simplified distributionally robust receiver. Simulations results are provided to show the effectiveness of our design by comparing with some existing well-known receivers.

We consider a homogeneous self-dual interior-point algorithm for solving multistage stochastic linear programs. The algorithm is particularly suitable for the so-called “scenario formulation” of the problem, whose constraint system consists of a large block-diagonal matrix together with a set of sparse nonanticipativity constraints. Due to this structure, the major computational work required by the homogeneous self-dual interior-point method can be split into three steps, each of which is highly decomposable. Numerical results on some randomly generated problems and a multistage production-planning problem are reported.

Abstract A quadratic black box is a two‐port network with each arc incurring a quadratic cost. This paper studies the properties of the minimum cost function of a quadratic black box with respect to an external flow input. An algorithm is presented with a numerical example for determining the subdifferential mapping of the minimum cost function. The algorithm can be applied to the piecewise quadratic case without change. If all data satisfy the commensurability condition, the algorithm terminates in a finite number of steps.