Introduction to the Special Issue on Autonomous Search

In the fall of 2007, we organized the first international workshop on Autonomous Search, colocated with the International Conference on Principles and Practice of Constraint Programming (CP’2007), which took place at Providence (RI). This event was an occasion to gather about 30 researchers from various research areas of constraint programming and operation research communities. Through this workshop, we wanted to promote works interested in the dynamic and autonomous adaptation of search procedures since recent progresses in the processing of combinatorial problems have demonstrated that search algorithms can become extremely efficient when they take advantage of unsuccessful attempts to drive their exploration. For instance in modern DPLLs, the collect of conflicts feeds the variable selection heuristic, and the quality of unit propagation can control the use of the restart strategy. We believe that a more principled and autonomous approach for search efficiency has to be started now in Constraint Programming. Therefore, following this workshop, we decided to provide, in this special issue, a short overview of some aspects of autonomous search through a selection of representative works. An Autonomous Search system has the ability to advantageously modify its internal components when exposed to changing external forces and opportunities. It corresponds to a particular case of adaptive systems whose objective is to improve its problem solving performance by adapting its search strategy to the problem at hand. Internal components correspond to the various algorithms involved in the search process heuristics, inferences, etc. External forces correspond to the evolving information collected during this search process. This information can be either directly collected on the problem or indirectly computed through the perceived efficiency of individual components. In this issue, we have selected three contributions to this new rising trend, covering two important families of combinatorial problems, namely constraint satisfaction problems (CSP) and Boolean satisfiability problems (SAT). From the methodological point of view, these works also address two main solving paradigms: complete methods, using exhaustive tree-based techniques and incomplete approaches, including evolutionary algorithms.