RULE-BASED ANNEALING (RA)
Heuristic knowledge (rules) are added to traditional Simulated Annealing
(SA) to generate a new configuration for a schedule table. Each rule is applied
probabilistically according to its selection ratio. Selection ratios of rules are
dynamically updated so that greedy rules are used more often when the ap-
pointed design time is short and random rearrangements are used more often
when there is ample time. This method produces a better solution within the
appointed time.
OUTLINE OF DEMONSTRATION (1)
(1) A Schedule table is a two-dimensional grid where each vertical slice
corresponds to a processing unit/ALU and each horizontal slice corre-
sponds to a time slot. Binomials (colored according to operation) are
placed in grid locations.
(2) Annealing temperature T and acceptance ratio 
in the case that
C (increase in cost) is 100.
(3) Total cost and each part of the total cost correspond to a schedule ta-
ble
(4) Total selection ratio of rules is classified by roles (to rearrange ran-
domly, to decrease ALU cost, and to decrease Time cost).
 Demonstration (1) |