Forward-Tracking: A Technique for Searching Beyond Failure

In many applications, such as decision support, negoti­ ation, planning, scheduling, etc., one needs to express re­ quirements that can only be partially satisfied. In order to express such requirements, we propose a technique called forward-tracking. Intuitively, forward-tracking is a kind o f dual o f chronological back-tracking: i f a program globally fails to find a solution, then a new execution is startedfrom a program point and a state ‘forw ard’ in the computation tree. This search technique is applied to the specific paradigm o f constraint logic programming, obtaining a powerful exten­ sion that preserves all the useful properties o f the original scheme. We report on the successful practical application o f forward-tracking to the evolutionary training o f (con­ strained) neural networks.