10A-D-1 Using Restricted Weight Update in Lagrangian Method for Optimization Problem(Room D International session)

抄録

In solving optimization problems, the most important thing is how effectively explore the solution space without being trapped by local minima. We have been developing optimization techniques which are based on Lagrangian method. In these methods, each constraint has weight, and all weights are updated in each step. This cause the ability of escaping from local minima. However we also found that updating the weights has undesirable side effect. In this paper we proposed a new method in which update of weights is done only in restricted steps. We apply this method to Software-MCU assignment problems and the experimental results show that we can improve the efficiency of the Lagrangian method further.