The optimization laboratory has a tradition of developing innovative platforms for optimization. The platforms implement the vision
which states that a combinatorial application is best described as a model and a search component. The model specifies what the solutions are and their overall quality in terms of constraints and objectives. The search expresses how to find solutions.
Comet is the award-winning research platform of the laboratory. It implements this vision for two different computational paradigms: constraint-based local search and constraint programming over finite domains. Comet models for these two paradigms are essentially similar and are expressed in rich modeling languages featuring logical and cardinality constraints, global/combinatorial constraints, numerical (possibly nonlinear) constraints, and vertical extensions for specific application areas. The search components are also expressed in a rich language for controlling graph- and tree- search explorations.
Comet also provides high-level abstractions for parallel computing to leverage the availability of commodity multi-processor or multi-core computers. These abstractions substantially reduce the distance between sequential and parallel implementations of an application.
Visualization is an important aspect of optimization research. Indeed, it is often hard to predict and understand the behavior of optimization algorithm. Comet provides a declarative language for specifying model-driven visualizations of various aspects of optimization algorithm.
The laboratory also aims at synthesizing the search of optimization applications, using the high-level structure exposed in models.
