Information Technology Reference
In-Depth Information
Table 3.1 Test problems.
Name
Inputs
Outputs
Function set
Two-bit full adder
5
3
XOR,MUX
Two-bit multiplier
4
4
AND,XOR
Even-three-parity
3
1
AND,OR,NAND,NOR
Even-four-parity
4
1
AND,OR,NAND,NOR
from the function or terminal sets. A specificity mutation can change both
dimensions of functionalities and specificity strengths, replacing the old value
with a randomly chosen new value.
DIGITAL CIRCUIT DESIGN
Enzyme GP has been applied to a number of problems in the domain of combi-
national logic design—the design of non-recurrent digital circuits. This section
discusses the performance of enzyme GP on the set of problems listed in Table
3.1. Parameter settings are recorded in Table 3.2. Results for enzyme GP using
uniform crossover, TR crossover, and no crossover are shown in Table 3.3.
Computational effort [16] measures the number of evaluations required for a
99% confidence of finding an optimal solution.
Table 3.3 shows that, in all but the even-four-parity problem, enzyme GP with
crossover performs better than enzyme GP with mutation alone. It also shows
that the computational effort using uniform crossover is less than using TR
crossover. Nevertheless, it is interesting to note that although uniform crossover
generally finds a solution in a higher proportion of runs, TR crossover generally
finds a solution earlier in a successful run. There are two apparent reasons that
uniform crossover has a higher success rate. First, genome length is bounded
so that all possible lengths are sufficient to contain an optimal solution. For TR
crossover this is not the case and, whereas solutions start with a viable length,
TR crossover is free to explore solutions that are shorter than the minimum
Table 3.2 Parameter settings.
Parameter
Value
Input bias (see Equation 1)
0.3
Number of specificities per enzyme
3
Distance limit for gene alignment in uniform c/o
1
Proportion of gene pairs recombined
15%
Transfer size range during TR crossover
1-5
Rate of specificity strength mutation
2%
Rate of functionality dimension mutation
15%
