Search

Much recent research effort has been directed to the development of efficient algorithms for solving minimax problems with theoretical convergence guarantees due to the relevance of these problems to a few emergent applications. In this paper, we propose a unified single-loop alternating gradient projection (AGP) algorithm for solving smooth nonconvex-(strongly) concave and (strongly) convex–nonconcave minimax problems. AGP employs simple gradient projection steps for updating the primal and dual variables alternatively at each iteration. We show that it can find an ε -stationary point of the objective function in O ε - 2 (resp. O ε - 4 ) iterations under nonconvex-strongly concave (resp. nonconvex–concave) setting. Moreover, its gradient complexity to obtain an ε -stationary point of the objective function is bounded by O ε - 2 (resp., O ε - 4 ) under the strongly convex–nonconcave (resp., convex–nonconcave) setting. To the best of our knowledge, this is the first time that a simple and unified single-loop algorithm is developed for solving both nonconvex-(strongly) concave and (strongly) convex–nonconcave minimax problems. Moreover, the complexity results for solving the latter (strongly) convex–nonconcave minimax problems have never been obtained before in the literature. Numerical results show the efficiency of the proposed AGP algorithm. Furthermore, we extend the AGP algorithm by presenting a block alternating proximal gradient (BAPG) algorithm for solving more general multi-block nonsmooth nonconvex-(strongly) concave and (strongly) convex–nonconcave minimax problems. We can similarly establish the gradient complexity of the proposed algorithm under these four different settings. [ABSTRACT FROM AUTHOR]

In this paper, we present difference of convex algorithms for solving bilevel programs in which the upper level objective functions are difference of convex functions, and the lower level programs are fully convex. This nontrivial class of bilevel programs provides a powerful modelling framework for dealing with applications arising from hyperparameter selection in machine learning. Thanks to the full convexity of the lower level program, the value function of the lower level program turns out to be convex and hence the bilevel program can be reformulated as a difference of convex bilevel program. We propose two algorithms for solving the reformulated difference of convex program and show their convergence to stationary points under very mild assumptions. Finally we conduct numerical experiments to a bilevel model of support vector machine classification. [ABSTRACT FROM AUTHOR]

The sparse nonlinear programming (SNP) problem has wide applications in signal and image processing, machine learning and finance, etc. However, the computational challenge posed by SNP has not yet been well resolved due to the nonconvex and discontinuous ℓ 0 -norm involved. In this paper, we resolve this numerical challenge by developing a fast Newton-type algorithm. As a theoretical cornerstone, we establish a first-order optimality condition for SNP based on the concept of strong β -Lagrangian stationarity via the Lagrangian function, and reformulate it as a system of nonlinear equations called the Lagrangian equations. The nonsingularity of the corresponding Jacobian is discussed, based on which the Lagrange–Newton algorithm (LNA) is then proposed. Under mild conditions, we establish the locally quadratic convergence and its iterative complexity estimation. To further demonstrate the efficiency and superiority of our proposed algorithm, we apply LNA to two specific problems arising from compressed sensing and sparse high-order portfolio selection, in which significant benefits accrue from the restricted Newton step. [ABSTRACT FROM AUTHOR]

In this paper, we devise new relaxations for composite functions, which improve the prevalent factorable relaxations, without introducing additional variables, by exploiting the inner-function structure. We outer-approximate inner-functions using arbitrary under- and over-estimators and then convexify the outer-function over a polytope P, which models the ordering relationships between the inner-functions and their estimators and utilizes bound information on the inner-functions as well as on the estimators. We show that there is a subset Q of P, with significantly simpler combinatorial structure, such that the separation problem of the graph of the outer-function over P is polynomially equivalent, via a fast combinatorial algorithm, to that of its graph over Q. We specialize our study to consider the product of two inner-functions with one non-trivial underestimator for each inner-function. For the corresponding polytope P, we show that there are eight valid inequalities besides the four McCormick inequalities, which improve the factorable relaxation. Finally, we show that our results generalize to simultaneous convexification of a vector of outer-functions. [ABSTRACT FROM AUTHOR]

Bilevel optimization problems have received a lot of attention in the last years and decades. Besides numerous theoretical developments there also evolved novel solution algorithms for mixed-integer linear bilevel problems and the most recent algorithms use branch-and-cut techniques from mixed-integer programming that are especially tailored for the bilevel context. In this paper, we consider MIQP-QP bilevel problems, i.e., models with a mixed-integer convex-quadratic upper level and a continuous convex-quadratic lower level. This setting allows for a strong-duality-based transformation of the lower level which yields, in general, an equivalent nonconvex single-level reformulation of the original bilevel problem. Under reasonable assumptions, we can derive both a multi- and a single-tree outer-approximation-based cutting-plane algorithm. We show finite termination and correctness of both methods and present extensive numerical results that illustrate the applicability of the approaches. It turns out that the proposed methods are capable of solving bilevel instances with several thousand variables and constraints and significantly outperform classical solution approaches. [ABSTRACT FROM AUTHOR]