Industrial problem solving using symbolic and subsymbolic AI

Effective and efficient decision making in industrial environments relies heavily on the optimization of complex problems to align them with specific objectives and constraints. However, modeling these mostly discrete optimization problems poses challenges, such as the selection of suitable modeling paradigms (e.g., mixed integer programming, constraint programming or heuristics) and the solvers used (e.g., Gurobi, IBM CP or Google or-tools). The industry is confronted with additional difficulties when adapting models, e.g. constraints may have to be adapted if business rules change. This in turn can result in high adaptation costs. New constraints can significantly affect both the solvability and runtime of algorithms, and symbolic solvers often face efficiency and scalability issues in industrial applications. This makes a flexible approach that abstracts problem modeling from the solution method highly desirable.
InProSSA aims to unify diverse solver paradigms under a common framework for addressing combinatorial optimization problems in industrial settings. An ambitious goal is to bridge the gap between symbolic and subsymbolic methods, integrating these contrasting approaches within a unified framework. In doing so, symbolic solvers (such as SAT solvers) will be utilized, and current subsymbolic methods, such as neural Monte Carlo Tree Search (MCTS), will be evaluated. The project is motivated by the principle that no single approach can solve all problems optimally, emphasizing the need for task-specific solution concepts (no-free-lunch theorem). The feasibility of this unified approach will be explored and evaluated within the framework of this exploratory project using an industrial use case. The focus is on exploring the integration of existing methods within this unified framework. All project outcomes, including articles, software, and data, will be made openly accessible to the public. Based on these results, research questions are derived and implemented in industrial research projects with company partners.
A local consortium of experts from the research areas of formal languages (Research Institute for Symbolic Computation), symbolic AI (Institute for Symbolic Artificial Intelligence), subsymbolic AI (RISC Software GmbH) and applications (RISC Software GmbH) is working on this task. The combination of these different competencies creates a new approach to solving industrial tasks.