Table 1 Lowest value achieved by various AL methods on synthetic benchmarks

From: Deep active optimization for complex systems

 

Ackley-20

Ackley-100

Rastrigin-20

Rastrigin-100

Rosenbrock-20

Rosenbrock-100

Schwefel-20

Griewank-20

Unit

1

1

×102

×103

×104

×104

×103

1

Maximum number of samples

1,600

2,800

1,000

2,000

6,300

10,500

1,000

1,000

Random

7.59 ± 0.17

9.23 ± 0.13

2.18 ± 0.15

1.47 ± 0.016

2.380 ± 0.119

64.60 ± 0.936

5.50 ± 0.11

233.1 ± 25.49

TuRBO5

0.37 ± 0.14

1.73 ± 0.18

0.52 ± 0.04

0.40 ± 0.034

0.003 ± 0.000

0.127 ± 0.066

2.84 ± 0.79

1.177 ± 0.049

LaMCTS

1.96 ± 0.75

5.05 ± 0.73

0.80 ± 0.30

0.82 ± 0.044

0.008 ± 0.005

0.652 ± 0.098

3.32 ± 0.33

0.956 ± 0.047

CMS-ES

0.75 ± 0.09

2.85 ± 0.04

0.78 ± 0.03

0.97 ± 0.017

0.006 ± 0.004

0.037 ± 0.004

5.28 ± 0.44

236.7 ± 45.85

Diff-Evo

6.43 ± 0.16

8.13 ± 0.19

1.88 ± 0.12

1.30 ± 0.032

0.797 ± 0.115

28.30 ± 2.690

5.10 ± 0.17

127.6 ± 12.25

DA

0.00 ± 0.00

3.28 ± 0.19

1.29 ± 0.06

0.53 ± 0.039

0.005 ± 0.003

0.908 ± 0.088

2.38 ± 0.39

1.252 ± 0.264

Shiwa

4.43 ± 0.07

5.78 ± 0.52

2.48 ± 0.02

1.19 ± 0.047

2.266 ± 0.146

0.240 ± 0.022

5.49 ± 0.32

0.175 ± 0.246

MCMC

0.00 ± 0.00

4.79 ± 0.16

0.89 ± 0.27

0.73 ± 0.038

0.011 ± 0.006

0.088 ± 0.036

2.11 ± 0.86

5.858 ± 8.782

DOO

7.17 ± 0.37

9.44 ± 0.09

2.22 ± 0.14

1.50 ± 0.044

1.640 ± 0.456

72.22 ± 2.700

5.56 ± 0.29

164.2 ± 21.41

SOO

7.75 ± 0.18

9.40 ± 0.17

2.24 ± 0.08

1.54 ± 0.027

2.760 ± 0.744

76.30 ± 2.700

2.89 ± 2.18

87.67 ± 4.048

VOO

2.44 ± 0.49

5.23 ± 0.17

1.03 ± 0.13

0.92 ± 0.028

0.006 ± 0.000

2.107 ± 0.324

5.38 ± 0.08

0.121 ± 0.091

DANTE

0.00 ± 0.00

0.00 ± 0.00

0.00 ± 0.00

0.00 ± 0.00

0.0003 ± 0.0005

0.002 ± 0.004

1.20 ± 0.49

0.000 ± 0.000

  1. Results are averaged over five trials, with ± indicating the s.d. The global optimum for these functions is 0. The bold font denotes the best results in this column.
Back to article page