Design Space Exploration of Parallel Architectures - Multi-objective Design Space Exploration of Multiprocessor SoC Architectures - page 187

0.05 0.05

0.04 0.04

0.03 0.03

0.02 0.02

0.01 0.01

0 0

100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000

Training set size Training set size

a b

Linear Regression Splines

0.05 0.01

0.04 0.008

0.03 0.006

0.02 0.004

0.01 0.002

0 0

100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000

Training set size Training set size

c d

Radial Basis Functions Kriging

0.01 0.01

0.008 0.008

0.006 0.006

0.004 0.004

0.002 0.002

0 0

100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000

Training set size Training set size

e f

Evolutionary Design Neural Networks

Fig. 8.11 Box plots of average normalized error versus training set size for logarithmic box-cox
transformation. a Linear regression, b Splines, c Radial basis functions, d Kriging, e Evolutionary
design, f Neural networks

for a specific use case depends on the desired accuracy and the constraints on the
prediction time. Referring to the use-case addressed in this section, however, we
discovered an exception to the previous empiric rule when, for λ

=

0, Kriging gave
a better accuracy than Evolutionary Design for every training set size, paired with a
lower computation time.