(200c) COLIN: Optimization Infrastructure for Hybrid Algorithms

While the optimization practitioner has a large library of optimization algorithms and approaches at their disposal, there are many challenging optimization problems (notably large-scale non-convex optimization and non-convex multi-objective optimization) that can not be reliably addressed by a single deterministic approach.  In these cases, an alternative approach is to turn to stochastic search routines like simulated annealing and genetic algorithms.  Past work has shown that it is possible to significantly improve the performance of these stochastic search routines by hybridizing these routines with local search capabilities in both structured [1] and unstructured [2] arrangements. However, the construction of these hybrid systems is a tedious and ad hoc process.

In this presentation, we present COLIN, a Common Optimization Library INterface for general optimization problems.  COLIN differs from other integrated optimization platforms (e.g., DAKOTA, GAMS, or AMPL) in that it does not require client problems and optimization algorithms to adhere to a strict (nominally MINLP) interface.  Instead, COLIN allows each problem and algorithm to implement only the interface that they directly support, using native data types of their choosing.  COLIN then provides the necessary functionality to dynamically resolve and transform both data types and entire problem representations as necessary to enable seamless communication between the problem and solver.  Similar transformations leverage concepts from Polymorphic Optimization [3] to support a global multi-solver, multi-problem annotated solution repository.  This flexible interface allows COLIN to support the rapid development and evaluation of structured and unstructured hybrid optimization algorithms.

[1] W.E. Hart and R.K. Belew. Optimization with genetic algorithm hybrids that use local search. In R.K. Belew and M. Mitchell, editors, Adaptive Individuals in Evolving Populations: Models and Algorithms, volume 26 of Santa Fe Institute studies in the science of complexity, pages 483-496. Addison-Wesley, 1996.

[2] J.D. Siirola, S. Hauan, and A.W. Westerberg. Computing Pareto fronts using distributed agents. Computers & Chemical Engineering, 29(1):113-126, 2004.

[3] J.D. Siirola and S. Hauan. Polymorphic Optimization. Computers & Chemical Engineering, 31(10):1312-1325, 2007.