A General Framework for Optimal Data-Driven Optimization
Speaker: Daniel Kuhn, EPFL
We propose a statistically optimal approach to construct data-driven decisions for stochastic optimization problems. Fundamentally, a data-driven decision is simply a function that maps the available training data to a feasible action. It can always be expressed as the minimizer of a surrogate optimization model constructed from the data. The quality of a data-driven decision is measured by its out-of-sample risk. An additional quality measure is its out-of-sample disappointment, which we define as the probability that the out-of-sample risk exceeds the optimal value of the surrogate optimization model. The crux of data-driven optimization is that the data-generating probability measure is unknown. An ideal data-driven decision should therefore minimize the out-of-sample risk simultaneously with respect to every conceivable probability measure (and thus in particular with respect to the unknown true measure). Unfortunately, such ideal data-driven decisions are generally unavailable. This prompts us to seek data-driven decisions that minimize the out-of-sample risk subject to an upper bound on the out-of-sample disappointment - again simultaneously with respect to every conceivable probability measure. We prove that such Pareto-dominant data-driven decisions exist under conditions that allow for interesting applications: the unknown data-generating probability measure must belong to a parametric ambiguity set, and the corresponding parameters must admit a sufficient statistic that satisfies a large deviation principle. If these conditions hold, we can further prove that the surrogate optimization model generating the optimal data-driven decision must be a distributionally robust optimization problem constructed from the sufficient statistic and the rate function of its large deviation principle. This shows that the optimal method for mapping data to decisions is, in a rigorous statistical sense, to solve a distributionally robust optimization model. Maybe surprisingly, this result holds irrespective of whether the original stochastic optimization problem is convex or not and holds even when the training data is non-i.i.d. As a byproduct, our analysis reveals how the structural properties of the data-generating stochastic process impact the shape of the ambiguity set underlying the optimal distributionally robust optimization model.
This is joint work with Tobias Sutter and Bart Van Parys
Exactness in SDP Relaxations of Quadratically Constrained Quadratic Programs
Speaker: Fatma Kılınç-Karzan, CMU
Abstract: Quadratically constrained quadratic programs (QCQPs) are a fundamental class of optimization problems. In a QCQP, we are asked to minimize a (possibly nonconvex) quadratic function subject to a number of (possibly nonconvex) quadratic constraints. Such problems arise naturally in many areas of operations research, computer science, and engineering. Although QCQPs are NP-hard to solve in general, they admit a natural family of tractable convex relaxations. In this talk, we will study the standard semidefinite program (SDP) relaxation for general QCQPs and examine when this relaxation is exact. By analyzing the "geometry" of the SDP relaxation, we will give both sufficient conditions and necessary conditions for different types of "exactness" conditions, and discuss a number of implications of these results for various applications.
This is joint work with Alex L. Wang and C.J. Argue
Convexification for Non-Convex Mixed-Integer Quadratic Programming
Speaker: Sam Burer, University of Iowa
Abstract: Convexification is an important technique used for solving non-convex mixed-integer quadratic programs. We discuss three recent convexification results for nonconvex quadratic programming over: (i) bounded (x1,x2,x3) with x1*x2 = x3; (ii) continuous (x1,x2) and binary (y1,y2) such that (0,0) <= (x1,x2) <= (y1,y2); and (iii) a ball intersected with a second-order cone. Although these structures may seem quite specialized, they appear as critical substructures in numerous applications. In addition to describing these three results, we survey the landscape---and the current research frontier---of convexification techniques in this area.
This talk reflects joint work with Kurt Anstreicher, Anders Eltved, and Kyungchan Park.