Computational Intelligence

A large number of problems require the optimization of multiple criteria. These crite­ ria are often non-commensurate and sometimes conflicting in nature making the task of optimization more difficult. In such problems, the task of creating a combined opti­ mization function is often not easy. Moreover, the decision procedure can be affected by the sensitivity of the solution space, and the trade-off is often non-linear. In real life we traditionally handle such problems by suggesting not one, but several non-dominated solutions. Finding a set of non-dominated solutions is also useful in multistaged opti­ mization problems, where the solution of one stage of optimization is passed on to the next stage. One classic example is that of circuit design, where high-level synthesis, logic synthesis and layout synthesis comprise important stages of optimization of the circuit. Passing a set of non-dominated partial solutions from one stage to the next typically ensures better global optimization. This book presents a new approach to multi-criteria optimization based on heuristic search techniques. Classical multicriteria optimization techniques rely on single criteria optimization algorithms, and hence we are either required to optimize one criterion at a time (under constraints on the others), or we are asked for a single scalar combined optimization function. On the other hand, the multiobjective search approach maps each optimization criterion onto a distinct dimension of a vector valued cost structure.