Use of Memory in Scatter Search

From a naive standpoint, virtually all heuristics other than complete randomization induce a pattern whose present state depends on the sequence of past states, and therefore incorporate an implicit form of “memory.” Given that the present is inherited from the past, the accumulation of previous choices is in a loose sense “remembered” by current choices. This sense is slightly more pronounced in the case of solution combination methods such as genetic algorithms and scatter search, where the mode of combination more clearly lends itself to transmitting features of selected past solutions to current solutions. Such an implicit memory, however, does not take a form normally viewed to be a hallmark of an intelligent memory construction. In particular, it uses no conscious design for recording the past and no purposeful manner of comparing previous states or transactions to those currently contemplated. By contrast, at an opposite end of the spectrum, procedures such as branch and bound and A* search use highly (and rigidly) structured forms of memory — forms that are organized to generate all nondominated solution alternatives with little or no duplication.