A causal framework for stochastic local search optimization algorithms

Despite the multitude of optimization algorithms available in the literature and the various approaches that study them, understanding the behaviour of an optimization algorithm and explaining its results are fundamental open questions in artificial intelligence and operations research. We argue that the body of available literature is already very rich, and the main obstacle to advancements towards answering those questions is its fragmentation.

In this work, we focus on stochastic local search algorithms, a broad class of methods to compute good quality suboptimal solutions in a short time. We propose a causal framework that relates the entities involved in the solution of an optimization problem. We demonstrate how this conceptual framework can be used to relate many approaches aimed at understanding how stochastic local search algorithms work, and how it can be utilized to address open problems, both theoretical and practical.