chapter6.ps

收集了遗传算法、进化计算、神经网络、模糊系统、人工生命、复杂适应系统等相关领域近期的参考论文和研究报告

4 10 Q
(or) 186.79 622.13 T
(not) 162.95 599.4 T
(d1) 165.6 578.84 T
(d2) 206.65 598.86 T
(or) 185.03 673.11 T
185.02 671.03 164.82 651.12 2 L
0.5 H
2 Z
N
192.19 671.14 209.74 654.25 2 L
N
(not) 154.12 644.85 T
(and) 206.21 646.05 T
209.74 644.51 193.85 629.9 2 L
N
217.68 644.51 233.57 629.9 2 L
N
158.54 642.89 149.82 626.77 2 L
N
(d1) 141.77 619.42 T
(not) 230.63 622.13 T
239.46 620.49 250.45 604.59 2 L
N
(d0) 246.82 596.15 T
193.85 620.16 209.74 605.55 2 L
N
185.9 620.16 170.01 605.55 2 L
N
170.01 598.64 170.01 586.07 2 L
N
328.66 639.82 348.77 633.74 328.5 628.24 328.58 634.03 4 Y
V
276.49 634.77 328.58 634.03 2 L
3 H
N
194.73 653.17 155 611.92 155 575.24 266.23 575.24 266.23 634.84 234.45 653.17 6 Y
1 H
0 Z
2 X
N
0 X
(or) 413.26 676.71 T
409.28 674.09 393.39 659.48 2 L
0.5 H
2 Z
N
425.17 674.09 441.06 659.48 2 L
N
(not) 384.56 651.43 T
389.86 650.28 389.86 635.67 2 L
N
(d1) 385.01 627.35 T
5 F
(newfunc) 426.93 651.66 T
(new) 351.71 598.78 T
(func) 371.15 598.78 T
6 14 Q
(\256) 395.03 598.78 T
4 10 Q
(\050and \050or \050not d1\051 d2\051) 412.34 598.78 T
7 F
(compression) 274.12 639.96 T
(operator) 280.5 622.43 T
4 F
(\050not d0\051\051) 427.16 586.43 T
108 90 540 702 C
0 0 612 792 C
FMENDPAGE
%%EndPage: "107" 5
%%Page: "108" 5
612 792 0 FMBEGINPAGE
108 63 540 702 R
7 X
0 K
V
108 711 540 720 R
V
0 12 Q
0 X
(108) 522.01 712 T
108 90 540 702 R
7 X
V
0 X
1.02 (Exactly how the randomly selected subtree is pruned to create a module\325) 126 475.19 P
1.02 (s de\336nition) 485.33 475.19 P
0.98 (depends on the compression method in use. The) 108 457.19 P
2 F
0.98 (depth compr) 349.33 457.19 P
0.98 (ession) 410.15 457.19 P
0 F
0.98 ( method extracts the) 440.14 457.19 P
0.56 (subtree from its root node and clips of) 108 439.19 P
0.56 (f any branch exceeding a randomly selected maxi-) 294.6 439.19 P
0.19 (mum depth. The maximum depth for each new module is generated by a uniform random) 108 421.19 P
-0.11 (variable over a user de\336ned range. Figure 24 shows the result of a compression when each) 108 403.19 P
0.52 (branch of the subtree is within the selected maximum depth. Here, compression removes) 108 385.19 P
0.31 (the entire subtree from the genotype, assigns it a unique module name) 108 367.19 P
1 F
0.31 (,) 447.18 367.19 P
0 F
0.31 (and de\336nes it as a) 453.49 367.19 P
0.86 (new LISP function with no parameters. A call to the newly de\336ned module replaces the) 108 349.19 P
1.28 (original subtree as shown in Figure 24. When GLiB evaluates the genotype\325) 108 331.19 P
1.28 (s \336tness, it) 486.46 331.19 P
2.03 (retrieves the compressed module\325) 108 313.19 P
2.03 (s de\336nition instead of directly executing the original) 274.96 313.19 P
(subtree.) 108 295.19 T
0.67 (When one or more subtree branches exceed the maximum depth, the de\336ned module) 126 265.19 P
0.37 (has a very dif) 108 247.19 P
0.37 (ferent de\336nition. Rather than extracting the entire subtree, only the portions) 173.85 247.19 P
0.55 (within the selected depth appear in the new module\325) 108 229.19 P
0.55 (s de\336nition. The branches exceeding) 362.25 229.19 P
1.36 (the depth are replaced with unique variable names. GLiB then de\336nes the new module) 108 211.19 P
-0.29 (with the variables de\336ned as parameters to the module. When GLiB replaces the subtree in) 108 193.19 P
1.22 (the genotype with the call to the module, the sections of the original subtree below the) 108 175.19 P
0.88 (selected depth become the parameter bindings. Figure 25 shows this instance of module) 108 157.19 P
(creation.) 108 139.19 T
108 90 540 702 C
124.88 483.19 523.12 702 C
140.62 501.19 500.62 528.19 R
7 X
0 K
V
3 12 Q
0 X
3.88 (Figure 25:) 140.62 520.19 P
2 F
3.88 (Depth compr) 204.37 520.19 P
3.88 (ession applied to a subtr) 270.76 520.19 P
3.88 (ee lar) 403.8 520.19 P
3.88 (ger than the) 434.89 520.19 P
(maximum depth. The module\325) 140.62 508.19 T
(s de\336nition has parameters as a r) 281.64 508.19 T
(esult.) 441.78 508.19 T
4 10 Q
(or) 212.82 670.64 T
(d0) 212.82 585.66 T
(d0) 181.55 556.03 T
218.03 669.2 197.18 652.21 2 L
0.5 H
0 Z
N
218.03 669.2 238.88 652.21 2 L
N
(not) 186.76 639.84 T
218.03 612.55 218.03 595.55 2 L
N
(or) 235.41 641 T
238.88 640.88 218.03 623.88 2 L
N
238.88 640.88 259.73 623.88 2 L
N
(d2) 254.52 608.32 T
282.49 681.46 285.79 693 289.1 681.46 285.79 681.46 4 Y
V
289.1 619.56 285.79 608.02 282.49 619.56 285.79 619.56 4 Y
V
285.79 681.46 285.79 619.56 2 L
1 H
N
(maxdepth) 0 -270 293.16 623.72 TF
(not) 212.82 612.68 T
191.97 640.88 171.12 623.88 2 L
N
(and) 160.7 612.68 T
165.91 612.55 145.06 595.55 2 L
0.5 H
N
165.91 612.55 186.76 595.55 2 L
N
(d1) 139.85 585.66 T
150.27 611.24 150.27 605.58 285.79 605.58 218.03 690.55 4 Y
1 H
N
5 F
(newfunc) 391.85 582.67 T
4 F
( \050) 432.38 582.67 T
7 F
(p1 p2 p3) 438.49 582.67 T
4 F
(\051) 477.39 582.67 T
6 14 Q
(\256) 483.22 582.67 T
4 10 Q
(\050or \050not \050and) 413.44 570.97 T
7 F
( p1 p2) 468.43 570.97 T
4 F
(\051\051) 496.22 570.97 T
(\050or \050not) 428.74 559.27 T
7 F
(p3) 463.73 559.27 T
4 F
(\051 d2\051\051\051) 474.85 559.27 T
(not) 181.54 584.35 T
186.76 582.92 186.76 565.92 2 L
0.5 H
N
371.15 628.34 391.34 622.56 371.15 616.77 371.15 622.56 4 Y
V
318.23 622.56 371.15 622.56 2 L
3 H
2 Z
N
7 F
(compression) 318.49 610.05 T
(depth) 330.65 632.06 T
5 F
(newfunc) 451.3 675.82 T
456.07 675 431.74 648.82 2 L
0.5 H
N
4 F
(d1) 423.63 639.82 T
464.17 675 464.17 648.82 2 L
N
(not) 456.07 639.82 T
(d0) 456.07 612.82 T
(d0) 488.5 639.82 T
472.28 675 496.61 648.82 2 L
N
464.17 639 464.17 621 2 L
N
108 90 540 702 C
0 0 612 792 C
FMENDPAGE
%%EndPage: "108" 6
%%Page: "109" 6
612 792 0 FMBEGINPAGE
108 63 540 702 R
7 X
0 K
V
108 711 540 720 R
V
0 12 Q
0 X
(109) 522.01 712 T
108 90 540 702 R
7 X
V
0 X
0.61 (Depth compression is but one method for extracting a module from an individual for) 126 442 P
1.01 (compression. An example of another modularization method is) 108 424 P
2 F
1.01 (leaf compr) 421.51 424 P
1.01 (ession) 473.7 424 P
0 F
1.01 (, which) 503.68 424 P
-0.12 (again selects a random node as the root of the subtree. However) 108 406 P
-0.12 (, in leaf compression, each) 412.59 406 P
0.22 (unique leaf is replaced with a unique parameter name, as shown in Figure 26. The advan-) 108 388 P
1.54 (tage of leaf compression over depth compression is the reuse of parameters within the) 108 370 P
2.27 (de\336ned module. Depth compression always produces a module in which a parameter) 108 352 P
-0.07 (appears only once. On the other hand, depth compression has the ability to construct mod-) 108 334 P
0.89 (ules from partial subtrees while leaf compression always compresses the full subtree. In) 108 316 P
0.89 (the experiments that follow) 108 298 P
0.89 (, each instance of compression randomly selects between the) 241.46 298 P
(depth compression and leaf compression methods with equal probability) 108 280 T
(.) 454.99 280 T
0.44 (Like the components of Hora\325) 126 250 P
0.44 (s watches, the modules formed by compression are sta-) 271.66 250 P
0.83 (ble and protected from dissessembly by other operators. Because the modules are repre-) 108 232 P
-0.19 (sented by single symbols, no crossover operation can af) 108 214 P
-0.19 (fect its de\336nition. The modules are) 373.73 214 P
1.51 (structurally preserved until an expansion operation replaces the symbol in the program) 108 196 P
0.59 (with its de\336nition. Schaf) 108 178 P
0.59 (fer and Morishima \0501987\051 discribe a dif) 227.49 178 P
0.59 (ferent method for evolv-) 420.66 178 P
1.29 (ing appropriate crossover points in a \336xed-length representation but with very dif) 108 160 P
1.29 (ferent) 512.03 160 P
(properties.) 108 142 T
107.5 450 540.5 702 C
127.79 474.19 509.68 516.25 R
7 X
0 K
V
3 12 Q
0 X
1.73 (Figure 26:) 127.79 508.25 P
2 F
1.73 (Leaf compr) 187.23 508.25 P
1.73 (ession applied to a selected subtr) 243.49 508.25 P
1.73 (ee.ote that identical) 411.28 508.25 P
1.83 (leaves ar) 127.79 496.25 P
1.83 (e r) 172.81 496.25 P
1.83 (epr) 187.19 496.25 P
1.83 (esented by the same parameter name in the de\336ned function.) 202.73 496.25 P
(Expansion is the r) 127.79 484.25 T
(everse of this pr) 214.3 484.25 T
(ocess.) 290.81 484.25 T
4 F
(or) 192.99 672.19 T
(d0) 193 579.86 T
(d0) 158.53 547.66 T
198.74 670.39 175.76 651.92 2 L
0.5 H
0 Z
N
198.74 670.39 221.72 651.92 2 L
N
(not) 164.27 638.73 T
198.74 608.83 198.74 590.37 2 L
N
(or) 217.9 639.99 T
221.72 639.61 198.74 621.14 2 L
N
221.72 639.61 244.7 621.14 2 L
N
(d2) 238.96 604.48 T
(not) 192.99 609.22 T
165.48 635.49 145.48 617.82 2 L
N
(and) 135.54 609.22 T
141.29 608.83 118.3 590.37 2 L
N
141.29 608.83 164.27 590.37 2 L
N
(d1) 112.56 579.86 T
5 F
(newfunc) 377.87 568.15 T
4 F
( \050) 426.51 568.15 T
7 F
(p1 p2 p3) 433.84 568.15 T
4 F
(\051) 480.52 568.15 T
6 F
(\256) 487.51 568.15 T
4 F
(\050or \050not \050and) 401.67 555.41 T
7 F
( p1 \050) 467.66 555.41 T
4 F
(not) 491.66 555.41 T
7 F
( p2) 508.33 555.41 T
4 F
(\051\051\051) 525 555.41 T
(\050or \050not) 418.54 542.7 T
7 F
(p) 460.53 542.7 T
4 F
(2\051) 467.2 542.7 T
7 F
(p3) 481.19 542.7 T
4 F
(\051\051) 494.53 542.7 T
(not) 158.52 578.44 T
164.27 576.64 164.27 558.17 2 L
N
352.38 623.4 363.91 620.09 352.38 616.79 352.38 620.09 4 Y
V
283.32 620.09 352.38 620.09 2 L
1 H
2 Z
N
7 F
(compression) 283.62 606.75 T
(leaf) 306.82 626.69 T
5 F
(newfunc) 426.58 655.31 T
439.34 650.53 412.53 622.09 2 L
0.5 H
N
4 F
(d1) 403.59 612.55 T
448.28 650.53 448.28 622.09 2 L
N
(d0) 439.01 606.75 T
(d2) 475.09 612.55 T
457.22 650.53 484.04 622.09 2 L
N
195.26 691.66 115.9 613.43 155.58 564.53 175.42 564.53 185.34 593.87 215.1 603.65 254.79 642.76 7 Y
N
0 0 612 792 C
FMENDPAGE
%%EndPage: "109" 7
%%Page: "110" 7
612 792 0 FMBEGINPAGE
108 63 540 702 R
7 X
0 K
V
108 711 540 720 R
V
0 12 Q
0 X
(1) 522.45 712 T
(10) 528.01 712 T
108 90 540 702 R
7 X
V
1 F
0 X
(6.2.2  Expansion of Compr) 108 694 T
(essed Modules) 244.39 694 T
0 F
0.54 (One drawback to the compression operator is that each compression removes genetic) 126 668 P
0.07 (material from the population and thus reduces the variability of the population. After only) 108 650 P
0.38 (a few generations of moderately applying the compression operator) 108 632 P
0.38 (, the available genetic) 434.29 632 P
0.93 (material in the population is insuf) 108 614 P
0.93 (\336cient to signi\336cantly improve the performance of the) 274.01 614 P
-0.15 (population. The loss of genetic diversity in a population leading to premature conver) 108 596 P
-0.15 (gence) 512.03 596 P
0.07 (to a suboptimal solution is a well known problem in genetic algorithms research \050De Jong) 108 578 P
0.74 (1975; Goldber) 108 560 P
0.74 (g 1989a; Davis 1991\051.) 178.15 560 P
0.74 (T) 289.47 560 P
0.74 (ypical solutions include altering the scaling of the) 295.95 560 P
0.05 (\336tness values or altering the frequency of crossovers and mutations.) 108 542 P
0.05 (Neither of these solu-) 436.26 542 P
(tions are appropriate for GLiB.) 108 524 T