The chemical casting model (CCM) is a computational model for solving difficult problems. The aim of the research on CCM is to establish a method of solving real-world problems, which are complex and open to the real world. Because complete and global infomation is usually impossible to obtain in real-world problem solving because of ``partiallily of information,'' we study local-information-based method. Computation that is based on local information and that causes non-local result, which may sometimes contain unestimable phenomena, is called emergent computation. CCM is a stochastic model for emergent computation.
CCM has been developed for computation using only local information. However, if the information is too local, it is not possible to find a solution, or to get into a globally stable state. Thus, controlling locality is important in local-information-based computation. Locality can be controlled easily in CCM by several methods: catalysts and rule composition. This controllability (or variability) of locality is an important feature of CCM.
CCM is based on a forward-chaining production system, such as systems written in OPS5 [For 81]. However, CCM differs from conventional forward-chaining production systems in several points: evaluation functions called LODs and randomized reactions.
The static structure and basic dynamics are explained.
Because of the above distinctive features, we believe that CCM is much more similar and analogical to chemical reaction systems than conventional production systems in AI. Thus, it is called chemical casting model. Production systems in AI are completely symbolic. However, CCM is more pattern-oriented because it is based on LODs that are numerical, and has certain similarity to neural systems. Because chemical reactions occur in parallel, it is very natural to make reactions occur asynchronously in parallel in CCM, and the parallelization is much easier in CCM than in conventional production systems.
A most recent (but unpublished) paper on CCM is ``Constraint Satisfaction by Parallel Optimization of Local Evaluation Functions with Annealing,'' but probably the best paper currently available for CCM is ``Stochastic Problem Solving by Local Computation based on Self-organization Paradigm'' (HICSS-27).