Title: A Domain Specific Language for Mathematical Programming Speaker: Joseph Young, Rice University Date/Time: Tuesday, March 11, 2008, 9:30 - 10:30 (MT) Location: CSRI/90 (NM) and 915/S145 (CA) Brief Abstract: The class of reliably solvable optimization problems has increased dramatically during the last ten years or so. These optimization problems include convex cone programs such as semidefinite programs and second-order cone programs. Although many of these problems have promising applications, tools that model these problems remain primitive. One difficulty in modeling these problems occurs during the formulation. Frequently, two or more equivalent formulations exist for a given model such that one formulation arises naturally, but the other one is more conducive to efficiently solving the problem. Since the process of transforming one formulation into the other can be difficult to an average practitioner, tools that automate this process are essential in making our problem-solving capability accessible. We present a new domain specific language for mathematical programming. The goal of this language is to represent a difficult optimization problem succinctly, analyze its properties, and then automate the transformation of one formulation to another. Our technical contribution to this area is the formal semantics of such a language, a static type system that gives information about the structure of the problem, and the properties of these transformations.CSRI POC: Scott Collis, (505) 284-1123 |