chapter3.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

(simply allowing a lar) 108 208 T
(ger amount of space for storing information about the search.) 209.39 208 T
2.1 (Recall from Equation 5, the expected number of population members at time) 126 178 P
2 F
2.1 (t+1) 522.57 178 P
0 F
(which contain feature) 108 160 T
3 F
(m) 214.91 160 T
0 F
( is given by:) 221.82 160 T
0 10 Q
(\050EQ 6\051) 512.53 120.97 T
204.27 101.21 416.25 138 C
2 12 Q
0 X
0 K
(m) 205.27 120.97 T
3 F
(m) 219.35 120.97 T
2 F
(t) 230.68 120.97 T
0 F
(1) 243.81 120.97 T
3 F
(+) 234.87 120.97 T
0 F
(\050) 214.64 121.97 T
(,) 225.54 121.97 T
(\051) 250.45 121.97 T
2 F
(n) 273.03 120.97 T
0 F
(1) 286.83 120.97 T
3 F
(e) 304.7 120.97 T
(-) 295.83 120.97 T
(m) 314.67 120.97 T
2 F
(t) 327.58 120.97 T
3 F
(,) 321.58 120.97 T
(\050) 310.68 120.97 T
(\051) 330.92 120.97 T
([) 281.73 120.97 T
(]) 335.52 120.97 T
(s) 357.2 120.97 T
2 F
(i) 372.23 120.97 T
(t) 381.57 120.97 T
3 F
(,) 375.57 120.97 T
(\050) 367.13 120.97 T
(\051) 385.51 120.97 T
2 9 Q
(i) 340.34 108.67 T
3 F
(L) 350 108.67 T
(m) 359.16 108.67 T
2 F
(t) 368.83 108.67 T
3 F
(,) 364.33 108.67 T
(\050) 356.16 108.67 T
(\051) 371.33 108.67 T
(\316) 343.84 108.67 T
3 18 Q
(\345) 342.79 117.77 T
3 12 Q
(=) 260.45 120.97 T
0 0 612 792 C
FMENDPAGE
%%EndPage: "48" 9
%%Page: "49" 9
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
(49) 528.01 712 T
108 90 540 702 R
7 X
V
0 X
-0.27 (where) 108 694 P
2 F
-0.27 (n) 140.03 694 P
0 F
-0.27 ( is the population size and) 146.03 694 P
3 F
-0.27 (e) 272.34 694 P
0 F
-0.27 (\050) 277.61 694 P
3 F
-0.27 (m) 281.6 694 P
0 F
-0.27 (,) 288.51 694 P
2 F
-0.27 (t) 294.24 694 P
0 F
-0.27 (\051 is the probability that the feature is disrupted dur-) 297.57 694 P
(ing reproduction. Equation 6 can be rewritten as follows:) 108 676 T
0 10 Q
(\050EQ 7\051) 512.53 634.99 T
(\050EQ 8\051) 512.53 578.05 T
0 12 Q
0.63 (where) 108 541.88 P
148.16 551.49 140.93 551.49 2 L
V
0.65 H
0 Z
N
3 F
0.63 (s) 140.93 541.88 P
0 F
0.63 (\050) 148.16 541.88 P
3 F
0.63 (m) 152.15 541.88 P
0 F
0.63 (,) 159.06 541.88 P
2 F
0.63 (t) 165.68 541.88 P
0 F
0.63 (\051 is the average probability of selection for a member of) 169.02 541.88 P
3 F
0.63 (L) 446.7 541.88 P
0 F
0.63 (\050) 454.93 541.88 P
3 F
0.63 (m) 458.92 541.88 P
2 F
0.63 (, t) 465.83 541.88 P
0 F
0.63 (\051. Notice that) 475.79 541.88 P
-0.3 (when) 108 523.88 P
3 F
-0.3 (m) 136.68 523.88 P
0 F
-0.3 ( is a schema,) 143.59 523.88 P
213.91 533.49 206.68 533.49 2 L
V
N
3 F
-0.3 (s) 206.68 523.88 P
0 F
-0.3 ( is de\336ned in accordance with \336tness proportionate reproduction and) 213.91 523.88 P
3 F
-0.13 (e) 108 505.88 P
0 F
-0.13 ( is de\336ned to adjust for crossover and point mutation then the lower bound expression for) 113.26 505.88 P
(the schema theorem of Holland \0501975\051 shown in Equation 3 is recovered.) 108 487.88 T
-0.03 (The interest in Equation 8 for abstract feature propagation stems from its characteriza-) 126 457.88 P
1.48 (tion of the properties of reproduction as the relative \336tness of the population members) 108 439.88 P
0.2 (with and without) 108 421.88 P
3 F
0.2 (m) 192.89 421.88 P
0 F
0.2 ( change. As long as) 199.8 421.88 P
304.63 431.49 297.4 431.49 2 L
V
N
3 F
0.2 (s) 297.4 421.88 P
0 F
0.2 (\050) 304.63 421.88 P
3 F
0.2 (m) 308.62 421.88 P
0 F
0.2 (,) 315.53 421.88 P
2 F
0.2 (t) 321.73 421.88 P
0 F
0.2 (\051 > 1) 325.06 421.88 P
2 F
0.2 (/\050n - n) 348.21 421.88 P
3 F
0.2 (e) 377.93 421.88 P
0 F
0.2 (\050) 383.19 421.88 P
3 F
0.2 (m) 387.18 421.88 P
0 F
0.2 (,) 394.09 421.88 P
2 F
0.2 (t) 400.29 421.88 P
0 F
0.2 (\051\051, the number of population) 403.62 421.88 P
-0.11 (members with the feature is likely to be lar) 108 403.88 P
-0.11 (ger in the next generation. On the other hand, if) 312.79 403.88 P
115.23 395.49 108 395.49 2 L
V
N
3 F
0.39 (s) 108 385.88 P
0 F
0.39 (\050) 115.23 385.88 P
3 F
0.39 (m) 119.23 385.88 P
0 F
0.39 (,) 126.13 385.88 P
2 F
0.39 (t) 132.52 385.88 P
0 F
0.39 (\051 < 1) 135.85 385.88 P
2 F
0.39 (/\050n - n) 159.38 385.88 P
3 F
0.39 (e) 189.47 385.88 P
0 F
0.39 (\050) 194.73 385.88 P
3 F
0.39 (m) 198.73 385.88 P
0 F
0.39 (,) 205.63 385.88 P
2 F
0.39 (t) 212.02 385.88 P
0 F
0.39 (\051\051 then the number of population members containing) 215.35 385.88 P
3 F
0.39 (m) 478.28 385.88 P
0 F
0.39 ( is likely to) 485.19 385.88 P
0.18 (decrease. In other words, as long as) 108 367.88 P
3 F
0.18 (m) 282.75 367.88 P
0 F
0.18 ( presents a suf) 289.66 367.88 P
0.18 (\336cient selection advantage to the sub-) 358.25 367.88 P
1.87 (population that contains it, additional population members tend to acquire the feature.) 108 349.88 P
0.55 (When) 108 331.88 P
3 F
0.55 (m) 140.19 331.88 P
0 F
0.55 ( is no longer an advantage, the feature is removed from the population automati-) 147.1 331.88 P
(cally by the natural dynamics of the evolutionary weak method.) 108 313.88 T
0.97 (Equation 8 characterizes the empirical power of the reproductive process used in all) 126 283.88 P
0.09 (evolutionary weak methods. It is this strength that separates evolutionary algorithms from) 108 265.88 P
0.3 (other search and optimization techniques and provides the adaptive capacity that separate) 108 247.88 P
0.28 (them from standard weak methods. Of equal importance is the generality and exploitabil-) 108 229.88 P
0.19 (ity of this reproductive process. For example, Davis \0501991\051 and Schwefel \0501981\051 describe) 108 211.88 P
1.05 (dif) 108 193.88 P
1.05 (ferent methods for evolving the parameters for manipulating population members for) 121.11 193.88 P
(two dif) 108 175.88 T
(ferent evolutionary weak methods.) 142.1 175.88 T
-0.11 (The necessity of a population for empirical credit assignment does not imply that stan-) 126 145.88 P
2.86 (dard weak methods that use populations, such as best-\336rst search and beam search,) 108 127.88 P
2.83 (embody empirical credit assignment: a population is necessary but not suf) 108 109.88 P
2.83 (\336cient. In) 491.2 109.88 P
179.83 615.23 440.7 654 C
2 12 Q
0 X
0 K
(m) 180.83 634.99 T
3 F
(m) 194.9 634.99 T
2 F
(t) 206.24 634.99 T
0 F
(1) 219.37 634.99 T
3 F
(+) 210.43 634.99 T
0 F
(\050) 190.19 635.99 T
(,) 201.1 635.99 T
(\051) 226.01 635.99 T
2 F
(n) 255.46 634.99 T
0 F
(1) 269.27 634.99 T
3 F
(e) 287.13 634.99 T
(-) 278.26 634.99 T
(m) 297.11 634.99 T
2 F
(t) 310.01 634.99 T
3 F
(,) 304.01 634.99 T
(\050) 293.11 634.99 T
(\051) 313.35 634.99 T
([) 264.17 634.99 T
(]) 317.95 634.99 T
2 F
(m) 323.77 643.57 T
3 F
(m) 340.24 643.57 T
2 F
(t) 353.14 643.57 T
3 F
(,) 347.15 643.57 T
(\050) 335.14 643.57 T
(\051) 357.09 643.57 T
2 F
(m) 323.77 627.37 T
3 F
(m) 340.24 627.37 T
2 F
(t) 353.14 627.37 T
3 F
(,) 347.15 627.37 T
(\050) 335.14 627.37 T
(\051) 357.09 627.37 T
(s) 381.64 634.99 T
2 F
(i) 396.68 634.99 T
(t) 406.01 634.99 T
3 F
(,) 400.01 634.99 T
(\050) 391.58 634.99 T
(\051) 409.95 634.99 T
2 9 Q
(i) 364.79 622.69 T
3 F
(L) 374.44 622.69 T
(m) 383.6 622.69 T
2 F
(t) 393.27 622.69 T
3 F
(,) 388.78 622.69 T
(\050) 380.61 622.69 T
(\051) 395.77 622.69 T
(\316) 368.28 622.69 T
3 18 Q
(\345) 367.24 631.78 T
3 12 Q
(=) 242.88 634.99 T
323.77 637.58 362.83 637.58 2 L
0.33 H
0 Z
N
0 0 612 792 C
201.84 573.88 418.68 591.23 C
2 12 Q
0 X
0 K
(m) 202.84 580.05 T
3 F
(m) 216.92 580.05 T
2 F
(t) 228.25 580.05 T
0 F
(1) 241.38 580.05 T
3 F
(+) 232.44 580.05 T
0 F
(\050) 212.21 581.05 T
(,) 223.11 581.05 T
(\051) 248.02 581.05 T
2 F
(n) 270.6 578.05 T
0 F
(1) 281.98 578.05 T
3 F
(e) 299.84 578.05 T
(-) 290.97 578.05 T
(m) 309.82 578.05 T
2 F
(t) 322.72 578.05 T
3 F
(,) 316.73 578.05 T
(\050) 305.82 578.05 T
(\051) 326.06 578.05 T
([) 276.88 578.05 T
(]) 330.66 578.05 T
2 F
(m) 339.79 578.05 T
3 F
(m) 354.38 578.05 T
2 F
(t) 367.29 578.05 T
3 F
(,) 361.29 578.05 T
(\050) 349.28 578.05 T
(\051) 371.23 578.05 T
(s) 379.8 578.05 T
(m) 392.96 578.05 T
2 F
(t) 405.87 578.05 T
3 F
(,) 399.87 578.05 T
(\050) 387.86 578.05 T
(\051) 409.81 578.05 T
(=) 258.02 578.05 T
380.8 586.04 387.03 586.04 2 L
0.33 H
0 Z
N
0 0 612 792 C
FMENDPAGE
%%EndPage: "49" 10
%%Page: "50" 10
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
(50) 528.01 712 T
108 90 540 702 R
7 X
V
0 X
1.4 (empirical credit assignment, the reproduction operators must map between members of) 108 694 P
1.22 (the problem space independent of task-speci\336c knowledge. Evolutionary weak methods) 108 676 P
0.29 (accomplish this by de\336ning representation-speci\336c operators for manipulation rather than) 108 658 P
0.29 (task-speci\336c operators as in standard weak/strong methods. T) 108 640 P
0.29 (o the extent that best \336rst or) 404.33 640 P
0.57 (beam search algorithms use representation speci\336c operators and not task-speci\336c opera-) 108 622 P
0.37 (tors, then these weak methods use a form of empirical credit assignment. However) 108 604 P
0.37 (, if the) 508.29 604 P
0.53 (operators that create new population members from old employ task-speci\336c knowledge,) 108 586 P
(as is intended in beam search and best-\336rst search, then the assignment of credit is strong.) 108 568 T
0.72 (In short, evolutionary weak methods gather a degree of information about the search) 126 538 P
0.5 (space by modifying the content of the population so that it contains abstract features that) 108 520 P
-0.15 (consistently associate with objects of better than average \336tness. Evolutionary weak meth-) 108 502 P
1.23 (ods become informed about the search space and the features important for solving the) 108 484 P
0.28 (problem during the run by manipulating the contents of the population. Informedness is a) 108 466 P
0.45 (quality consistently missing from the standard weak methods described above. The types) 108 448 P
0.69 (of features preserved and passed on to subsequent populations are determined chie\337y by) 108 430 P
0.61 (the reproduction operators and the method of selecting which objects in a population are) 108 412 P
(used to create the next.) 108 394 T
1 F
(3.2  Emergent Intelligence) 108 358 T
0 F
-0.21 (Empirical credit assignment uses the \336tness function as an integral part of its computa-) 126 334 P
1.75 (tion. Unlike stronger forms of credit assignment, empirical credit assignment does not) 108 316 P
0.1 (deduce the validity of representational components but empirically determines an abstract) 108 298 P
2.51 (feature\325) 108 280 P
2.51 (s validity though modi\336cation and retesting. This sets up a strong interaction) 144.63 280 P
1.12 (between the evolutionary weak method and the task environment. Through this interac-) 108 262 P
0.14 (tion, solution constraints that are implicit within the task environment become manifested) 108 244 P
2.39 (in the population as they are discovered. V) 108 226 P
2.39 (iewing the system at the knowledge-level) 329.19 226 P
1.12 (\050Newell 1982\051, an observer would deduce that the evolutionary weak method possesses) 108 208 P
0.94 (task-speci\336c knowledge that lead to the solution\325) 108 190 P
0.94 (s structure. Given that there is no task-) 348.19 190 P
-0.13 (speci\336c knowledge in the problem solver) 108 172 P
-0.13 (, the interaction of the evolutionary weak method) 304.41 172 P
0.32 (with the task environment allows knowledge to emer) 108 154 P
0.32 (ge at the knowledge-level that is not) 364.19 154 P
(available to the problem solver as explicit knowledge.) 108 136 T
FMENDPAGE
%%EndPage: "50" 11
%%Page: "51" 11
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
(51) 528.01 712 T
108 90 540 702 R
7 X
V
1 F
0 X
(3.2.1  Emergence and Emergent Computation) 108 694 T
0 F
0.14 (Emer) 126 668 P
0.14 (gent intelligence is a direct generalization from the properties of empirical credit) 151.76 668 P
2.23 (assignment and evolutionary weak methods that exploits the task environment during) 108 650 P
2.41 (problem solving. Emer) 108 632 P
2.41 (gent intelligence suggests that interaction of a simple problem) 222.88 632 P
0.58 (solver with a task environment can remove the need for task-speci\336c knowledge internal) 108 614 P
1.74 (to the problem solver) 108 596 P
1.74 (. Before a more adequate explanation of emer) 214.82 596 P
1.74 (gent intellige