chapter0.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

(REPRESENT) 140.4 590 T
(A) 211.49 590 T
(TIONS) 219.26 590 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 260.86 590 T
( 104) 511.51 590 T
0 F
(6.1  Problem Decomposition) 140.4 568 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 281.85 568 T
( 104) 511.51 568 T
(6.2  The Genetic Library Builder) 140.4 554 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 302.83 554 T
( 106) 511.51 554 T
(6.2.1  Compression of Modules) 162 540 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 317.83 540 T
( 107) 511.51 540 T
(6.2.2  Expansion of Compressed Modules) 162 524 T
( . . . . . . . . . . . . . . . . . . . . . . .) 368.8 524 T
( 1) 511.73 524 T
(10) 520.28 524 T
(6.2.3  Non-monotonic Formation of Hierarchical Decompositions) 162 508 T
( . . . .) 482.73 508 T
( 1) 511.73 508 T
(10) 520.28 508 T
(6.2.4  Evaluation of Modules) 162 492 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 305.83 492 T
( 1) 511.95 492 T
(1) 520.5 492 T
(1) 526.05 492 T
(6.2.5  The Emer) 162 476 T
(gence of High-Level Representations) 239.39 476 T
( . . . . . . . . . . . . . .) 422.77 476 T
( 1) 511.73 476 T
(12) 520.28 476 T
(6.2.6  T) 162 460 T
(ask-Speci\336c Feature Acquisition) 198.47 460 T
(. . . . . . . . . . . . . . . . . . . . . . . . .) 359.8 460 T
( 1) 511.73 460 T
(13) 520.28 460 T
(6.3  Generality of Emer) 140.4 444 T
(gent Modules) 253.76 444 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 323.82 444 T
( 1) 511.73 444 T
(14) 520.28 444 T
(6.4  Comparison to Koza\325) 140.4 430 T
(s ADFs) 263.99 430 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 305.83 430 T
( 1) 511.73 430 T
(14) 520.28 430 T
(6.5  The Emer) 140.4 416 T
(gence of Environment Speci\336c Modules) 208.8 416 T
(. . . . . . . . . . . . . . . . .) 407.78 416 T
( 1) 511.73 416 T
(15) 520.28 416 T
(6.5.1  T) 162 402 T
(ower of Hanoi Experiments) 198.47 402 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 335.82 402 T
( 1) 511.73 402 T
(15) 520.28 402 T
(6.5.2  Experiments with T) 162 386 T
(ic T) 286.84 386 T
(ac T) 304.98 386 T
(oe) 325.12 386 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . .) 341.81 386 T
( 1) 511.73 386 T
(17) 520.28 386 T
(6.6  Discussion) 140.4 370 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 218.88 370 T
( 123) 511.51 370 T
(6.7  Conclusion) 140.4 356 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 221.88 356 T
( 123) 511.51 356 T
2 F
(VII) 113.4 332 T
(EMERGENT GOAL-DIRECTED BEHA) 140.4 332 T
(VIOR) 351.42 332 T
( . . . . . . . . . . . . . . . . . . . .) 386.79 332 T
( 125) 511.51 332 T
0 F
(7.1  Competitive Learning) 140.4 310 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 272.85 310 T
( 126) 511.51 310 T
(7.2  Competitive Learning in Evolutionary Algorithms) 140.4 296 T
(. . . . . . . . . . . . . . . . .) 407.78 296 T
( 128) 511.51 296 T
(7.2.1  Standard Fitness Functions and Competitive Selection) 162 282 T
(. . . . . . . . .) 455.75 282 T
( 129) 511.51 282 T
(7.2.2  Competitive Fitness Functions) 162 266 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . .) 341.81 266 T
( 130) 511.51 266 T
(7.2.2.1  Axelrod\325) 194.4 250 T
(s Full Competitive Model) 276.02 250 T
( . . . . . . . . . . . . . . . . .) 404.78 250 T
( 130) 511.51 250 T
(7.2.2.2  Hillis\325 Bipartite Competitive Model) 194.4 234 T
( . . . . . . . . . . . . . . . .) 410.77 234 T
( 132) 511.51 234 T
(7.3  T) 140.4 218 T
(ournament Fitness) 167.88 218 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 260.86 218 T
( 134) 511.51 218 T
(7.4  Experiments) 140.4 204 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 227.88 204 T
( 136) 511.51 204 T
(7.5  Results) 140.4 190 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 200.89 190 T
( 138) 511.51 190 T
(7.6  Discussion) 140.4 176 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 218.88 176 T
( 141) 511.51 176 T
(7.7  Conclusion) 140.4 162 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 221.88 162 T
( 142) 511.51 162 T
2 F
(VIII) 111.07 138 T
(SUMMAR) 140.4 138 T
(Y AND CONCLUSIONS) 195.27 138 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 329.82 138 T
( 144) 511.51 138 T
(BIBLIOGRAPHY) 108 94 T
( 152) 519.01 94 T
FMENDPAGE
%%EndPage: "ix" 12
%%Page: "x" 12
612 792 0 FMBEGINPAGE
108 72 540 81 R
7 X
0 K
V
0 12 Q
0 X
(x) 321 73 T
108 90 540 648 R
7 X
V
2 F
0 X
(LIST OF T) 275.79 640 T
(ABLES) 332.89 640 T
(T) 108 582 T
(ABLE) 115.11 582 T
(P) 507.57 582 T
(AGE) 514.01 582 T
0 F
(1.) 121.5 558 T
(Differences between artificial intelligence and emergent intelligence.) 144 558 T
(. . . . . .) 479.74 558 T
( 59) 514.5 558 T
(2.) 121.5 538 T
(Characterizations of the seven languages from Tomita \0501982\051.) 144 538 T
( . . . . . . . . . . .) 446.75 538 T
( 81) 514.5 538 T
(3.) 121.5 518 T
(Positive and negative examples for the seven languages investigated in) 144 518 T
(Tomita \0501982\051.) 144 504 T
1 F
( e) 216.62 504 T
0 F
( signifies the null string.) 224.88 504 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 344.81 504 T
( 83) 514.5 504 T
(4.) 121.5 482 T
-0.4 (Results from Watrous and Kuhn \0501992\051 for the seven languages shown in) 144 482 P
(Table 3.) 144 468 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 188.9 468 T
( 84) 514.5 468 T
(5.) 121.5 446 T
-0.28 (Experimental results for GNARL on the seven languages shown in Table) 144 446 P
(3. Units are identical for the same measures in Table 4.) 144 432 T
(. . . . . . . . . . . . . . . . .) 413.77 432 T
( 85) 514.5 432 T
(6.) 121.5 410 T
-0.65 (Speed-up results for FSA induction for 4 runs with and without freeze and) 144 410 P
(unfreeze.) 144 396 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 191.9 396 T
( 101) 511.51 396 T
(7.) 121.5 374 T
(Table showing results of competition experiments.) 144 374 T
( . . . . . . . . . . . . . . . . . . .) 392.78 374 T
( 138) 511.51 374 T
FMENDPAGE
%%EndPage: "x" 13
%%Page: "xi" 13
612 792 0 FMBEGINPAGE
108 72 540 81 R
7 X
0 K
V
0 12 Q
0 X
(xi) 319.33 73 T
108 90 540 648 R
7 X
V
2 F
0 X
(LIST OF FIGURES) 272.35 640 T
(FIGURE) 108 582 T
(P) 507.57 582 T
(AGE) 514.01 582 T
0 F
(1.) 121.5 558 T
(Comparison of the effectiveness of weak and strong methods over the) 144 558 T
-0.26 (variety of tasks that can be attempted. Strong methods are specialized for) 144 544 P
(a set of tasks and more effective while weak methods are general across) 144 530 T
(tasks but less effective.) 144 516 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 260.86 516 T
(18) 516 516 T
(2.) 121.5 494 T
(Algorithm for Generate and Test weak method.) 144 494 T
( . . . . . . . . . . . . . . . . . . . . . . .) 374.79 494 T
( 19) 514.5 494 T
(3.) 121.5 474 T
(Algorithm for hill climbing weak method. the switch function switches) 144 474 T
(the values of its arguments so that each has the others value.) 144 460 T
(. . . . . . . . . . . . .) 437.76 460 T
(20) 516 460 T
(4.) 121.5 438 T
(Recursive algorithm for depth-first search. O is the set of operators that) 144 438 T
-0.64 (defines the connectivity of the search space. The connectivity of the space) 144 424 P
(is assumed to be a tree.) 144 410 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 260.86 410 T
(21) 516 410 T
(5.) 121.5 388 T
-0.24 (Algorithm for beam search.) 144 388 P
3 F
-0.24 (obj-queue) 278.61 388 P
0 F
-0.24 ( and) 326.57 388 P
3 F
-0.24 (next-queue) 349.41 388 P
0 F
-0.24 ( are standard LIFO) 402.03 388 P
(queues with standard operators) 144 374 T
3 F
(head) 295.89 374 T
0 F
(,) 319.2 374 T
3 F
(push) 325.2 374 T
0 F
( and) 347.85 374 T
3 F
(pop) 371.17 374 T
0 F
(. The sort routine is) 389.16 374 T
-0.55 (assumed to use Tester to determine the ordering so that the objects closest) 144 360 P
-0.57 (to being solutions are at the front of the queue. The space is assumed to be) 144 346 P
(a tree.) 144 332 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 176.9 332 T
( 22) 514.5 332 T
(6.) 121.5 310 T
(Generic algorithm for evolutionary algorithms.) 144 310 T
( . . . . . . . . . . . . . . . . . . . . . . .) 374.79 310 T
( 26) 514.5 310 T
(7.) 121.5 290 T
(The separation of structure and behavior. \050a\051 Emphasis on acquiring) 144 290 T
-0.34 (structure results in creating offspring that are structurally similar but may) 144 276 P
(be behaviorally incomparable. \050b\051 Emphasis on acquiring behaviors) 144 262 T
(might require operators that associate dissimilar structures.) 144 248 T
(. . . . . . . . . . . . . .) 431.76 248 T
( 28) 514.5 248 T
(8.) 121.5 226 T
-0.61 (Operators used by genetic algorithms. \050a\051 single point crossover; \050b\051 point) 144 226 P
(mutation; \050c\051 inversion.) 144 212 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 260.86 212 T
( 32) 514.5 212 T
(9.) 121.5 190 T
-0.12 (Structure of a perceptron. The binary retina feeds into a set of masks that) 144 190 P
(essentially multiply their inputs. These binary variables are then) 144 176 T
-0.6 (multiplied by the weights and summed. If the sum is larger than the stored) 144 162 P
(threshold, then the concept represented by the preceptron appears in the) 144 148 T
(retina.) 144 134 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 179.9 134 T
( 66) 514.5 134 T
FMENDPAGE
%%EndPage: "xi" 14
%%Page: "xii" 14
612 792 0 FMBEGINPAGE
108 72 540 81 R
7 X
0 K
V
0 12 Q
0 X
(xii) 317.67 73 T
108 90 540 648 R
7 X
V
0 X
(10.) 118.5 640 T
(Standard connectionist architecture.) 144 640 T
3 F
(x1) 319.2 640 T
0 F
( and) 330.52 640 T
3 F
(x2) 353.84 640 T
0 F
( are inputs to the network) 365.16 640 T
(and are set by the environment. Their activation is multiplied by the) 144 626 T
(weight on the associated connection. A hidden or output node sums all) 144 612 T
-0.14 (inputs, subtracts the threshold which is shown as a value inside the node,) 144 598 P
-0.76 (and applies a sigmoid activation function to determine output. The state of) 144 584 P
-0.8 (the output node is the The above network computes the exclusive-or of the) 144 570 P
(inputs.) 144 556 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 179.9 556 T
( 67) 514.5 556 T
(11.) 118.5 534 T
(The dual representation scheme used in genetic algorithms. The) 144 534 T
-0.93 (interpretation function maps between the elements in recombination space) 144 520 P
-0 (on which the search is performed and the subset of structures that can be) 144 506 P
(evaluated as potential task solutions.) 144 492 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 323.82 492 T
( 70) 514.5 492 T
(12.) 118.5 470 T
(The competing conventions problem \050Schaffer, Whitley and Eshelman) 144 470 T
(1992\051. Bit strings A and B map to structurally and computationally) 144 456 T
(equivalent networks that assign the hidden units in different orders.) 144 442 T
(Because the bit strings are distinct, crossover is likely to produce an) 144 428 T
-0.19 (offspring that contains multiple copies of the same hidden node, yielding) 144 414 P
(a network with less computational ability than either parent.) 144 400 T
(. . . . . . . . . . . . .) 437.76 400 T
( 72) 514.5 400 T
(13.) 118.5 378 T
(The evolutionary programming approach to modeling evolution. Unlike) 144 378 T
(genetic algorithm, evolutionary programs perform search in the space of) 144 364 T
(networks. Offspring created by mutation remain within a locus of) 144 350 T
(similarity to their parents.) 144 336 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 272.85 336 T
( 74) 514.5 336 T
(14.) 118.5 314 T
(Sample initial network. The number of input nodes \050min\051 and number of) 144 314 T
-0.32 (output nodes \050mout\051 is fixed for a given task. The presence of a bias node) 144 300 P
-0.31 (\050b = 0 or 1\051 as well as the maximum number of hidden units \050hmax\051 is set) 144 286 P
(by the user. The initial connectivity is chosen randomly \050see text\051. The) 144 272 T
(disconnected hidden node does not affect this particular network\325s) 144 258 T
(computation, but is available as a resource for structural mutations.) 144 244 T
( . . . . . . .) 470.74 244 T
( 77) 514.5 244 T
(15.) 118.5 222 T
(The number of network evaluations required to learn the seven data sets) 144 222 T
(of Table 3. GNARL \050using both SAE and SSE fitness measures\051) 144 208 T
-0.73 (compared to the average number of evaluations for the five runs described) 144 194 P
(in Watrous and Kuhn \0501992\051.) 144 180 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 290.84 180 T
( 86) 514.5 180 T
(16.) 118.5 158 T
(Percentage accuracy of evolved networks on languages in Table 3.) 144 158 T
(GNARL \050using SAE and SSE fitness measures\051 compared to average) 144 144 T
(accuracy of the five runs in Watrous and Kuhn \0501992\051.) 144 130 T
( . . . . . . . . . . . . . . . . .) 410.77 130 T
( 87) 514.5 130 T
FMENDPAGE
%%EndPage: "xii" 15
%%Page: "xiii" 15
612 792 0 FMBEGINPAGE
108 72 540 81 R
7 X
0 K
V
0 12 Q
0 X
(xiii) 316 73 T
108 90 540 648 R
7 X
V
0 X
(17.) 118.5 640 T
-0.76 (The ant problem. Black squares indicate food while grey squares show the) 144 640 P
(shortest path. The trail is connected initially, but becomes progressively) 144 626 T
(more difficult to follow. The underlying 2-d grid is toroidal, so that) 144 612 T
-0.16 (position \322A\323 is the first break in the trail \320 it is simple to reach this point.) 144 598 P
-0.91 (Positions \322B\323 and \322C\323 indicate the only two positions along the trail where) 144 584 P
(the ant discovered in run 1 behaves differently from the 5-state FSA of) 144 570 T
(Jefferson et al. \0501992\051 \050see Figure 20\051.) 144 556 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 332.82 556 T
( 88) 514.5 556 T
(18.) 118.5 534 T
(The semantics of the I/O units for the ant network. The first input node) 144 534 T
-0.07 (denotes the presence of food in the square directly in front of the ant; the) 144 520 P
(second denotes the absence of food in this same square. This particular) 144 506 T
(network finds 42 pieces of food before spinning endlessly in place at) 144 492 T
(position P, illustrating a very deep local minimum in the search space.) 144 478 T
(. . . . .) 485.73 478 T
( 89) 514.5 478 T
(19.) 118.5 456 T
-0.42 (The Tracker Task, first run. \050a\051 The best network in the initial population.) 144 456 P
(Nodes 0 and 1 are input, nodes 5-8 are output, and nodes 2-4 are hidden) 144 442 T
-0.21 (nodes. \050b\051 Network induced by GNARL after 2090 generations. Forward) 144 428 P
(links are dashed; bidirectional links and loops are solid. The light gray) 144 414 T
(connection between nodes 8 and 13 is the sole backlink. This network) 144 400 T
(clears the trail in 319 epochs.) 144 386 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 287.84 386 T
( 90) 514.5 386 T
(20.) 118.5 364 T
(FSA hand-crafted for the Tracker task by Jefferson et al. \0501992\051. Food is) 144 364 T
-0.19 (indicated by 1, no food is indicated by 0. This simple system implements) 144 350 P
-0.31 (the strategy \322move forward if there is food in front of you, otherwise turn) 144 336 P
(right four times, looking for food. If food is found while turning, pursue) 144 322 T
(it, otherwise, move forward one step and repeat.\323 This simple FSA) 144 308 T
-0.28 (traverses the entire trail in 314 steps, and gets a score of 81 in the allotted) 144 294 P
(200 time steps.) 144 280 T
(. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 221.88 280 T
( 91) 514.5 280 T
(21.) 118.5 258 T
(Limit behavior of the network that clears the trail in 319 steps. Graphs) 144 258 T
(show the state of the output units Move, Right, Left. \050a\051 Fixed point) 144 244 T
(attractor that results for sequence of 500 \322food\323 signals; \050b\051 Limit cycle) 144 230 T
-0.39 (attractor that results when a sequence of 500 \322no food\323 signals is given to) 144 216 P
-0.77 (network; \050c\051 All states visited while traversing the trail; \050d\051 The path of the) 144 202 P
-0.19 (ant on an empty grid. The z axis represents time. Note that) 144 188 P
3 F
-0.19 (x) 425.49 188 P
0 F
-0.19 ( is fixed, and) 430.81 188 P
3 F
-0.65 (y) 144 174 P
0 F
-0.65 ( increases monotonically at a fixed rate. The large jumps in) 149.32 174 P
3 F
-0.65 (y) 429.32 174 P
0 F
-0.65 ( position are) 434.65 174 P
(artifacts of the toroidal grid.) 144 160 T
( . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) 284.84 160 T
( 92) 514.5 160 T
FMENDPAGE
%%EndPage: "xiii" 16
%%Page: "xiv" 16
612