System-Level Design:Genetic Algorithms
Genetic Algorithms
Genetic algorithms are becoming an important tool for solving the highly nonlinear problems related to system-level synthesis. The use of genetic algorithms in optimization is well discussed by Michalewicz [61] where formulations for problems such as bin packing, processor scheduling, traveling salesman, and system partitioning are outlined.
Research works involving the use of genetic algorithms to system-level synthesis problems are starting to be published, as for example the results of
• Hou et al. [62]—scheduling of tasks in a homogeneous multiprocessor without communication costs.
• Wang et al. [63]—scheduling of tasks in heterogeneous multiprocessors with communication costs, but not allowing cost versus performance trade-off, i.e., all processors have the same cost.
• Ravikumar and Gupta [64]—mapping of tasks into a reconfigurable homogeneous array processor without communication costs.
• Tirat-Gefen and Parker [65]—a genetic algorithm for design of application-specific heterogeneous multiprocessors (ASHM) with nonnegligible communications costs specified by a nonperiodic task-flow graph representing both control and data flow.
• Tirat-Gefen [54]—introduced a full-set of genetic algorithms for system-level design of ASHMs incorporating new design features such as imprecise computation and probabilistic design.
Comments
Post a Comment