Heuristic search strategy

For some designs, exact exploration of the design space may be prohibitive due to its size. Hebe also supports heuristic strategies to explore the design space, where the resulting implementation is no longer guaranteed to be optimal. 

The ranking is based on evaluating the candidate resource bindings with respect to a set of cost criteria. Three cost criteria are supported interconnect. 

The decision of whether one binding is better than another dq)ends on the relative importance of each criteria. This is determined by a weight associated with each criterion. The total cost CtotiP) binding / is givCT as follows, where C, t, Carea, and C, trial denote the interconnect, area, and soialization costs, respectively. 

0) — Wjnt Cfnl I Warea C area I serial Gterial 

The values Wi t, Warea, and w,trial are weight factors corresponding to each cost criterion. The designer can also specify cut-off thresholds for each cost criterion such that a binding with costs exceeding the thresholds will not be considered for synthesis. These three cost criteria are described in detail below. 

---

Pearl, J. "Heuristics-Intelligent Search Strategies for Computer Problem-Solving", Addison-Wesley Publ. Co., Reading, MA, 1984. 

For small numbers of decisions, say up to 10-15, total enumeration is recommended, regardless of relaxation quality. As the number of binary decisions increases, models subdivide into three regions. In the outermost, relaxations are too poor to assist in analysis. The only practical strategy is some form of heuristic search lacking even a bound on the suboptimality of results obtained. 

Up to version 3.2 of the EROS system the selection of the chemically interesting reactions was based on heuristic rules for finding the appropriate bonds that were to be submitted to the reaction schemes. Furthermore, for each reaction the enthalpy was calculated and this value was used for selecting the best pathways. Clearly, in the selection process a distinction has to be made between a forward and a retrosynthetic search. Thus, the value of the heat of reaction is used with different sign and weight in the selection process for the two types of search strategies. 

To solve a problem of this first class is to perform a tree search. Ponton and Donaldson use heuristics to select each next match and find only one solution, often a good one but not always. Pho and Lapidus propose a total enumeration scheme, but, for very large problems (10 streams), suggest a fallible lookahead strategy to eliminate branches. Lee, Masso and Rudd,... 

---