N. J. Radcliffe. Genetic set recombination and its application to neural network topology optimisation. Edinburgh Parallel Computing Centre Technical Report EPCC-TR-91-21, 1991.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....issues to be addressed by research in the field [Montana 1989] Radcliffe 1990] Apart from an enormous enlargement of the search space arising from permutational redundancy there are other potentially serious problems which can arise from naive representations. This point is made forcefully in [Radcliffe 1991b] the danger is exemplified by the case where two equivalent networks (identical up to a relabelling of hidden units) can be recombined to produce a child which is not equivalent to them. This is a phenomenon which is not seen in conventional genetic search (for example, single parameter ....

N. J. Radcliffe. Genetic set recombination and its application to neural network topology optimisation. Edinburgh Parallel Computing Centre Technical Report EPCC-TR-91-21, 1991.

....numbers are never changed. Thus, the historical origin of every gene in the system is known throughout evolution. The historical markings give NEAT a powerful new capability, effectively solving the problem of competing conventions for disparate topologies (the Holy Grail in neuroevolution [14]) The system now knows exactly which genes match up with which (figure 3) Genes that do not match are either disjoint or excess, depending on whether they occur within or outside the range of the other parent s innovation numbers. When crossing over, the genes in both genomes with the same ....

N. J Radcliffe. Genetic set recombination and its application to neural network topology optimisation. Neural computing and applications, 1(1):67--90, 1993.

....Mills, Graham, 21, 22] Mitchell, Melanie, 35, 36, 37, 53, 54, 60, 61] Montana, David J. 68] Moriarty, David, 67] Muntean, Traian, 86, 87] Naillon, Martine, 24] Napliotis, Nicholas, 56] Ostermeier, Andreas, 69] Perkins, Sonya, 21] Petridis, V. 8, 18] Radcliffe, Nicholas J. [9, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81] Romaniuk, Steve G. 82] Santib a nez Koref, Ivan, 29, 30] Saravanan, N. 4] Schwefel, Hans Paul, 83] Silva, A. De, 22] Surry, Patrick, 9] Tackett, Walter Alden, 84] Talbi, El Ghazali, 85, 86, 87, 88, 89, 90] Tarroux, Philippe, 10] Thornton, Chris, 15] Voigt, Hans Michael, ....

....13 3.9 Annual index: 1957 1994 The following table gives references to the contributions published during the period 1957 1994. 1990, 38, 70] 1991, 39, 53, 62, 71, 72, 73, 74, 75, 85, 86, 87] 1992, 19, 20, 31, 35, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 66, 76, 77, 78, 79, 88, 89] 1993, [21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 32, 33, 34, 4, 36, 37, 50, 51, 52, 54, 55, 56, 57, 58, 59, 60, 61, 63, 64, 65, 67, 68, 69, 80, 81, 82, 83, 84, 90] 1994, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16] 1995, 17, 18] 14 Chapter 4 Permuted title index The words of the titles of the articles are shown in the next table arranged in alphabetical order. The most common words have been excluded. The key word is shown by a disk (ffl) in the title field ....

Nicholas J. Radcliffe. Genetic set recombination and its application to neural network topology optimization. Neural Computing and Applications, 1(1):67--90, 1993. (also as [73]; available via anonymous ftp cite ftp.epcc.ed.ac.uk directory /pub/tr/91 file tr9121.ps.Z) ga:Radcliffe93a.

....list contains references to all papers published as technical reports. The list is arranged in alphabetical order by the name of the institute. Aristotle University of Thessaloniki, 7] Bolt Beranek and Newman, 68] California Institute of Technology, 23] Edinburgh Parallel Computing Centre, [9, 70, 72, 73, 75, 78, 79] Florida Atlantic University, 4] International Computer Science Institute, 33, 34] International Computer Science Institute (ICSI) 31] Iowa State University, 55] Mitsubishi Electric Research Laboratories, 26] National University of Singapore, 82] Santa Fe Institute, 36, 60, 61] ....

....Mills, Graham, 21, 22] Mitchell, Melanie, 35, 36, 37, 53, 54, 60, 61] Montana, David J. 68] Moriarty, David, 67] Muntean, Traian, 86, 87] Naillon, Martine, 24] Napliotis, Nicholas, 56] Ostermeier, Andreas, 69] Perkins, Sonya, 21] Petridis, V. 8, 18] Radcliffe, Nicholas J. [9, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81] Romaniuk, Steve G. 82] Santib a nez Koref, Ivan, 29, 30] Saravanan, N. 4] Schwefel, Hans Paul, 83] Silva, A. De, 22] Surry, Patrick, 9] Tackett, Walter Alden, 84] Talbi, El Ghazali, 85, 86, 87, 88, 89, 90] Tarroux, Philippe, 10] Thornton, Chris, 15] Voigt, Hans Michael, ....

[Article contains additional citation context not shown here]

Nicholas J. Radcliffe. Genetic set recombination and its application to neural network topology optimization. Technical Report TR-91-21, Edinburgh Parallel Computing Centre, 1991. (also as [80]; available via anonymous ftp cite ftp.epcc.ed.ac.uk directory /pub/tr/91 file tr9121.ps.Z) ga:Radcliffe91b.

....is not necessarily more efficient than direct encoding methods. Section 4.3.3) We now turn to several specific problems with TWEANNs and address each in turn. 2. 2 Competing Conventions One of the main problems for NE is the Competing Conventions Problem, also known as the Permutations Problem (Radcliffe 1993). Competing conventions means having more than one way to express a solution to a weight optimization problem with a neural network. When genomes representing the same solution do not have the same encoding, crossover is likely to produce damaged offspring. Figure 1 depicts the problem for a ....

....or even genomes of different sizes. Because TWEANNs do not satisfy strict constraints on the kinds of topologies they produce, proposed solutions to the competing conventions problem for fixed or constrained topology networks such as non redundant genetic encoding (Thierens 1996) do not apply. Radcliffe (1993) goes as far as calling an integrated scheme combining connectivity and weights the Holy Grail in this area. Although some TWEANNs such as PDGP (Pujol and Poli 1998) have attempted to address the problem by assuming that subnetworks represent functional units that can be recombined, different ....

Radcliffe, N. J. (1993). Genetic set recombination and its application to neural network topology optimisation. Neural computing and applications, 1(1):67--90.

....Technol. I, Sci. Eng. Japan) 565] Methods of Information in Medicine, 901] Midwest Symp Circuits Syst, 273] Mini Micro Syst. China) 438] Modell Simul Mater Sci Eng, 284] Network: Computation in Neural Systems, 836] Neural Computat. Appl. 558] Neural Computing and Applications, [887] Neural Computing Applications, 31] Neural Netw. World (Czech Republic) 390] Neural Network Review, 977] Neural Network World, 246, 309, 865] Neural Networks, 41, 49, 90, 154, 567, 799, 946] Neural Parallel Sci. Comput, 30] Neural Process. 36] Neural Process. Lett. ....

....L. 881] Prasanth, Ravi K. 944] Pratt, P. 76] Price, J. E. 456] Prieto, A. 153, 198, 293, 513, 865, 866] Protzel, P. 416] Puigjaner, L. 329] Pyeatt, Larry, 303, 341] Qiang, Wang, 457] Qizhi, Zhang, 374] Rabelo, L. C. 255] Rabelo, Luis, 207] Radcliffe, Nicholas J. [885, 886, 887] Ragg, T. 381] Rajasekaran, S. 382, 541] Rajroop, P. 57] Ramasamy, J. V. 382] RamBabu, P. 110] Ranson, Aaron L. 888] Rastogi, Ravi, 28] Ray, K. S. 356] Reed, R. 883] Reeves, Colin R. 889, 890, 891, 892] Rehder, J. 229] Reidys, C. 248] Reilly, K. D. 351] ....

[Article contains additional citation context not shown here]

Nicholas J. Radcliffe. Genetic set recombination and its application to neural network topology optimization. Neural Computing and Applications, 1(1):67--90, 1993. (also as [886]; available via anonymous ftp site ftp.epcc.ed.ac.uk directory /pub/tr/91 file tr9121.ps.Z) ga:Radcliffe93a.

....of the institute. Academy of Sciences of the USSR, 636] Carnegie Mellon University, 787] Colorado State University, 945, 948, 949, 950] Deutsches Elektronen Synchrotron, 628] Ecole Normale Superiore, 772] Ecole Normale Sup erieure de Lyon, 716] Edinburgh Parallel Computing Centre, [886] Honeywell Corporate Systems, 727, 729] Institute of Psychology CNR, 834] Iowa State University, 150] LASPP FER, 662] NIBS Pte Ltd. 35] 14 Genetic algorithms and neural networks National Research Counsil (C. N. R. 271, 372] National University of Singapore, 896] Naval Command, ....

....L. 881] Prasanth, Ravi K. 944] Pratt, P. 76] Price, J. E. 456] Prieto, A. 153, 198, 293, 513, 865, 866] Protzel, P. 416] Puigjaner, L. 329] Pyeatt, Larry, 303, 341] Qiang, Wang, 457] Qizhi, Zhang, 374] Rabelo, L. C. 255] Rabelo, Luis, 207] Radcliffe, Nicholas J. [885, 886, 887] Ragg, T. 381] Rajasekaran, S. 382, 541] Rajroop, P. 57] Ramasamy, J. V. 382] RamBabu, P. 110] Ranson, Aaron L. 888] Rastogi, Ravi, 28] Ray, K. S. 356] Reed, R. 883] Reeves, Colin R. 889, 890, 891, 892] Rehder, J. 229] Reidys, C. 248] Reilly, K. D. 351] ....

[Article contains additional citation context not shown here]

Nicholas J. Radcliffe. Genetic set recombination and its application to neural network topology optimization. Technical Report TR-91-21, Edinburgh Parallel Computing Centre, 1991. (also as [887]; available via anonymous ftp site ftp.epcc.ed.ac.uk directory /pub/tr/91 file tr9121.ps.Z) ga:Radcliffe91b.

....Visualization and Computer Animation, 700] Knowledge Based Systems (UK) 164] Kybernetes, 549] Meas Sci Technol, 181] Microelectron. J. 176] Microprocessors and Microsystems, 283] Microprocessors and Microsystems (UK) 91] Nature, 481, 482, 484] Neural Computing and Applications, [683] Neural Computing Applications, 30] Neural Networks (USA) 306] New Scientist, 604, 673] Nuclear Engineer, 663] Online and CD ROM Review, 40] Parallel Computing, 589] Pattern Recognition Letters, 343] Power Syst. Technol. China) 285] Proc Inst Mech Eng Part B J Eng Manuf, ....

....344, 359, 364, 369, 370, 372, 373, 665, 666] Prager, Richard, 16] Probert, Penelope, 58, 331] Prosser, P. 185] Prugel Bennett, A. 51] Purchase, G. 427] Purvis, A. 645, 646] Pye, C. J. 266] Quick, R. J. 423] Quinn, L. 353, 668] Rabelo, L. C. 324] Radcliffe, Nicholas J. [18, 80, 93, 134, 148, 11, 437, 671, 672, 673, 674, 675, 676, 677, 678, 679, 680, 681, 682, 683, 684] Rafiq, Y. 413] Rakowski, Tomasz, 295] Rana, A. S. 268] Rastogi, M. 318] Rattray, M. 51] Rayne, C. M. 79] Raywardsmith, V. J. 310] Rayward Smith, V. J. 329, 423] Rayward Smith, Vic J. 105, 54, 81, 233, 541, 542] Redmill, D. W. 286] Reeves, Colin R. 82, 135, 149, ....

[Article contains additional citation context not shown here]

Nicholas J. Radcliffe. Genetic set recombination and its application to neural network topology optimization. Neural Computing and Applications, 1(1):67--90, 1993. (also as [676]; available via anonymous ftp cite ftp.epcc.ed.ac.uk directory /pub/tr/91 file tr9121.ps.Z ) ga:Radcliffe93a.

....Wales, 618] total 8 thesis in 8 schools 4.4 Report series The following list contains references to all papers published as technical reports. The list is arranged in alphabetical order by the name of the institute. Coventry University, 222] EPRI, 160] Edinburgh Parallel Computing Centre, [18, 672, 675, 676, 678, 681, 682, 707, 708] IBM, 698] Loughborough University of Technology, 667] Patents 11 Plymouth Engineering Design Centre, 502, 503] The University of Sheffield, 515] University College London, 601] University of Cambridge, 472] University of Durham, 639] University of East Anglia, 138, 539] ....

....344, 359, 364, 369, 370, 372, 373, 665, 666] Prager, Richard, 16] Probert, Penelope, 58, 331] Prosser, P. 185] Prugel Bennett, A. 51] Purchase, G. 427] Purvis, A. 645, 646] Pye, C. J. 266] Quick, R. J. 423] Quinn, L. 353, 668] Rabelo, L. C. 324] Radcliffe, Nicholas J. [18, 80, 93, 134, 148, 11, 437, 671, 672, 673, 674, 675, 676, 677, 678, 679, 680, 681, 682, 683, 684] Rafiq, Y. 413] Rakowski, Tomasz, 295] Rana, A. S. 268] Rastogi, M. 318] Rattray, M. 51] Rayne, C. M. 79] Raywardsmith, V. J. 310] Rayward Smith, V. J. 329, 423] Rayward Smith, Vic J. 105, 54, 81, 233, 541, 542] Redmill, D. W. 286] Reeves, Colin R. 82, 135, 149, ....

[Article contains additional citation context not shown here]

Nicholas J. Radcliffe. Genetic set recombination and its application to neural network topology optimization. Technical Report TR-91-21, Edinburgh Parallel Computing Centre, 1991. (also as [683]; available via anonymous ftp cite ftp.epcc.ed.ac.uk directory /pub/tr/91 file tr9121.ps.Z ) ga:Radcliffe91b.

....[190] Kikai Gijutsu Kenkyusho Shoho, 997] Machine Learning, 313, 386, 391, 516, 1065] Machine Learning Journal, 216] Mech. Syst. Signal Process. UK) 193] Methods of Information in Medicine, 881] Microprocessing and Microprogramming, 1073] Neural Computing and Applications, [847] Neural Network World, 802, 1004] New Electronics (UK) 799] New Scientist, 271, 661] Nippon Kikai Gakkai Ronbunshu A Hen, 760, 762, 1080] Nippon Kikai Gakkai Ronbunshu C Hen, 339, 340, 563] NKK Technical Report (Japan) 320] Opt. Mem. Neural Netw. USA) 452] Optical Engineering, ....

....Pulat, Simin, 501] Punch, W. F. 256] Purdin, T. D. M. 826] Purvis, A. 774, 775] Putnam, Jeffrey, 827] Qi, Xiaofeng, 829, 830] Qian, Ahihang, 831] Qiu, Yuping, 206] Quafafou, Mohamed, 742] Quinn, L. 832] Quintana, Chris, 599] Rabelo, Luis, 35] Radcliffe, Nicholas J. [847, 848] Rahman, Saifur, 849] Rahmani, Adel Torkaman, 768, 769] Rajeev, S. 850, 851] Ralston, Patricia A. S. 423] Ramanathan, Parameswaran, 403] Ramsey, Connie Loggia, 387] Rangel, Naykiavick, 94, 95] Rankin, Richard Patrick, 853] Rankin, R. 852] Ranson, Aaron L. 854] Rao, B. B. ....

[Article contains additional citation context not shown here]

Nicholas J. Radcliffe. Genetic set recombination and its application to neural network topology optimization. Neural Computing and Applications, 1(1):67--90, 1993. (also as [?]; available via anonymous ftp cite ftp.epcc.ed.ac.uk directory /pub/tr/91 file tr9121.ps.Z ) ga:Radcliffe93a.

.... of Chemical Physics, 75] Journal of Computer Aided Molecular Design, 36] Journal of Molecular Biology, 84] Lettre du Transputer et des Calculateurs Distribu es, 373] Machine Learning, 316] Mathware Soft Computing, 201] Molecular Simulations, 96] Neural Computing and Applications, [364] Nucleic Acids Research, 54] Pattern Recognition Letters, 42] Scientific Computing World, 71] Technique et Science Informatique TSI, 371] Transactions of the Institute of Electronics, Information and Communication Engineers A (Japan) 41] Transactions of the Institute of Electronics, ....

....[270] Phon Amnuaisuk, Somnuk, 95] Postaire, Jack G erard, 279] Potter, Mitchell A. 82] Preux, Philippe, 167] Probert, Penelope, 124, 188] Proenca, Luis Miguel, 21, 24, 26] Punch, William F. 55, 84] Punch, III, William F. 49, 79] Pyylampi, Tero, 101, 294] Radcliffe, Nicholas J. [111, 354, 355, 356, 357, 358, 359, 360, 361, 362, 363, 364, 365] Ranito, J. V. 21] Rasanen, Petri, 121] Raymer, Michael L. 55, 84] Reeves, Colin R. 126, 138, 215] Romaniuk, Steve G. 366] Rooij, A. J. F. Van, 64] Rosca, Justinian, 281] Rost, Ursula, 272] Rowe, Jon, 169, 229] Ryan, Conor, 287] Ryynanen, Matti, 127, 230, 240, 291] Saarinen, ....

[Article contains additional citation context not shown here]

Nicholas J. Radcliffe. Genetic set recombination and its application to neural network topology optimization. Neural Computing and Applications, 1(1):67--90, 1993. (also as [357]; ftp.epcc.ed.ac.uk: /pub/tr/91/ tr9121.ps.Z) Key: ga:Radcliffe93a.

....by the name of the institute. Akademie der Wissenschaft der DDR, 306] Argonne National Laboratory, 136] Aristotle University of Thessaloniki, 107] Bolt Beranek and Newman, 350] California Institute of Technology, 299] Carnegie Mellon University, 11, 44] Edinburgh Parallel Computing Centre, [111, 354, 356, 357, 359, 362, 363] Florida Atlantic University, 4] International Computer Science Institute (ICSI) 310, 312, 313] Iowa State University, 337] Leiden University, 194] Michigan State University, 48] Mitsubishi Electric Corp. 303] Mitsubishi Electric Research Laboratories, 302] National University of ....

....[270] Phon Amnuaisuk, Somnuk, 95] Postaire, Jack G erard, 279] Potter, Mitchell A. 82] Preux, Philippe, 167] Probert, Penelope, 124, 188] Proenca, Luis Miguel, 21, 24, 26] Punch, William F. 55, 84] Punch, III, William F. 49, 79] Pyylampi, Tero, 101, 294] Radcliffe, Nicholas J. [111, 354, 355, 356, 357, 358, 359, 360, 361, 362, 363, 364, 365] Ranito, J. V. 21] Rasanen, Petri, 121] Raymer, Michael L. 55, 84] Reeves, Colin R. 126, 138, 215] Romaniuk, Steve G. 366] Rooij, A. J. F. Van, 64] Rosca, Justinian, 281] Rost, Ursula, 272] Rowe, Jon, 169, 229] Ryan, Conor, 287] Ryynanen, Matti, 127, 230, 240, 291] Saarinen, ....

[Article contains additional citation context not shown here]

Nicholas J. Radcliffe. Genetic set recombination and its application to neural network topology optimization. Technical Report TR-91-21, Edinburgh Parallel Computing Centre, 1991. (also as [364]; ftp.epcc.ed.ac.uk: /pub/tr/91/ tr9121.ps.Z) Key: ga:Radcliffe91b.

....an appropriate coding scheme and suitable genetic operators for the given problem. Unfortunately, finding the best coding and genetic operators is not generally a trivial task, and is, in fact, a major stumbling block which plagues efforts to optimise neural networks with GAs (e.g. Hancock 1992, Radcliffe 1991). In this section we describe our coding scheme for the bumptree. The bumptree classifier is coded as a chromosome of fixed length, built up of 30 blocks of real valued genes, as shown in figure 1. 1 2 3 4 29 30 5 6 7 8 9 28 1 2 3 4 5 6 7 8 9 10 11 12 13 14 0.367 0.421 0.789 0.001 0.200 ....

....leading to a situation analogous to two normal people producing a child with two left feet. This effect reduces the GA s ability to discover and combine good genetic building blocks. This kind of redundancy in genetic codings for neural networks has been called the permutation problem (e.g. Radcliffe, 1991), and has proved a formidable obstacle in attempts to optimise the MLP genetically. Consider using a GA to optimise a fully connected MLP with a single hidden layer of n nodes, coded genetically as an ordered list. As the functionality of the MLP is independent of any order in the hidden units, ....

Radcliffe, N., 1991, Genetic Set Recombination and its Application to Neural Network Topology Optimisation, Report No. EPCC-TR-91-21, University of Edinburgh.

....that, in the neural context, this is the task for which continuous EAs are most naturally suited. It is difficult to find applications in which GAs (or other EAs, for that matter) have outperformed DBMs for supervised training of feed forward neural networks [25] It has been pointed out [20] that this task is inherently hard for algorithms that rely heavily on the recombination of potential solutions. In addition, the training times can become too costly, even worse than that for DBMs. The emergence of evolutionary techniques using different selection schemes and a direct ....

....135 5 8 11 14 17 20 23 26 29 32 35 38 41 44 47 50 Average Best Figure 6: Results as a function of truncation parameter . Left: EIR (ffi = 0:35) Right: EIR (range ffi ) Each point is the result of NRuns=20 runs. begins to decline, for EIR (range ffi ) the best region appears very soon, in [8, 20], peaking at 11 and 17. These zones are marked in the plots of Fig. 6 (left and right) by two vertical bars. In between (in [20, 26] there is a transition zone, where performance for both is in the [118, 122] mark. More precisely, the crossing lies in [23, 26] where both operators are in the ....

[Article contains additional citation context not shown here]

Radcliffe, N.J. Genetic set recombination and its application to neural network topology optimization. Technical Report EPCC-TR-91-21. Edinburgh Parallel Computing Centre. Univ. of Edinburgh, Scotland, 1991. 16

.... 6= g(10) 6= g(11) 6= g(00) In contrast, a representation is said to exhibit degeneracy if more than one chromosome can represent the same solution, i.e. if the genotype phenotype mapping is non injective) Degeneracy is widely, but not universally, perceived as detrimental to genetic search (Radcliffe, 1993). 4.3.4 Linkage Considerations The linkage of a collection of genes refers to its likelihood of being transmitted en masse from one parent to a child under the action of recombination. Under the action of N point crossover, for instance, a group of genes is said to be tightly linked if they ....

....limits. 8. 2 Scaling Properties of Discretised Representations Although previous work on forma analysis has concentrated primarily on finite search domains, such as combinatorial problems like the travelling sales rep problem (Radcliffe Surry, 1994b) neural network topology optimisation (Radcliffe, 1993), multiobjective pipeline optimisation (Surry et al. 1995) and so forth, some initial work was done on continuous domains (Radcliffe, 1991a, 1991b) In this chapter we extend these ideas by considering a limiting sequence of discrete representations. These results are used to define two formal ....

N. J. Radcliffe, 1993. Genetic set recombination and its application to neural network topology optimisation. Neural Computing and Applications, 1(1):67--90.

....thinking about the TSP and population approaches for it. If a sentence could condense it would be the following Whitley [133] and Radcliffe [109] have both argued that in tackling the TSP it is essential to focus attention on edges rather than nodes. where we did not add the emphasis (see Ref. [110]) Obviously, this concept is not new for researchers in Operations Research and Graph Theory, though it seems that those coming from GA community may have just discovered it. It is also evident that it is definitely the case for the TSP since the objective function is a sum over edges and there ....

N.J. Radcliffe, Genetic Set Recombination and its Application to Neural Network Topology Optimization, Neural Computing and Applications, 1 (1992) 1.

....Unfortunately there are still no general principles to guide the design of a structure appropriate to a given problem. Possible approaches include weight pruning [9] weight sharing [11] and constructive algorithms [2, 3] A number of authors have suggested using GAs to specify the structure, eg [7, 10, 1, 8, 12]. The GA specifies an architecture, which is then trained on the target problem using Backprop or some equivalent. The test score is returned to the GA as the evaluation of the design. The problems addressed have usually been small, such as Exclusive Or, because of the excessive cpu demand of the ....

N. Radcliffe. Genetic set recombination and its application to neural network topology optimisation. Neural computing and applications, 1:67--90, 1993.

....competing conventions in parallel. Braun and Weisbrod [11] tried to prevent permuted internal representations by making long connections less probable than short connections and thus preferring for each functional mapping the structural mapping with the shortest connection lengths. Radcliffe [57] suggested a matching recombination operator based on the pattern of connections of the hidden units. This and some similar recombination operators are critically compared in [24] see also [72] for some experiments) Montana and Davis [52] matched hidden units before crossover by their responses ....

N. Radcliffe. Genetic set recombination and its application to neural network topology optimization. Technical Report EPCC-TR-91-21, Edinburgh Parallel Computing Centre, University of Edinburgh, 1991.

....model. Whitley et al. (1991) report successful results for training the weights of nets by GA, using what they term a genetic hill climber, which is essentially mutation driven. This is indicated by a decrease in the number of evaluations required as the population size is reduced to one. Recently, Radcliffe (1993) has suggested a matching recombination operator that at2 tempts to overcome the permutation problem. Preliminary tests on this operator, and an extension proposed by the author, indicated that both tended to perform less well than a simple form of crossover (Hancock, 1992c) This paper explores ....

....strings. Ideally, we should like a method of identifying the equivalent units prior to crossover. Montana and Davis (1989) suggested such a method for use in the context of training net weights by GA. They matched hidden units by their responses to a number of trial inputs applied to the net. Radcliffe (1993) has suggested a method that applies to net structure definition. This treats net specifications as a multiset, where the elements are the possible hidden unit connectivities. The hidden layer is a multiset, because it is possible to have more than one copy of a unit with given connectivity (in ....

Radcliffe, N. 1993. Genetic set recombination and its application to neural network topology optimisation. Neural computing and applications, 1, 67--90.

No context found.

, 1992.

....if the formae are believed to be the appropriate ones for capturing the performance regularities in the search space and it is accepted that respect and assortment contribute to effective genetic search. Defining an operator which either respects or assorts an arbitrary set of formae is trivial (Radcliffe, 1990) Radcliffe (1992c) has also shown that a set of parameterised operators can be introduced which assort arbitrary formae and for which the degree of violation of respect can be controlled. In essence, the greater the degree of violation of respect that is permitted, the more thorough is the assortment. Clearly ....

....of the members of E. A spanning set E is then said to constitute an orthogonal basis for Psi (Radcliffe, 1991b) if given any choice of equivalence class for each of the relations in E, a solution in S exists in their mutual intersection. E is said to constitute an independent basis for Psi (Radcliffe, 1992b) if none of the members of E can be constructed by intersecting other members of Psi. Orthogonality is the stronger condition, and implies independence (Radcliffe, 1992a) The equivalence relations in an orthogonal basis E for a set Psi of equivalence relations which covers S are called basic ....

[Article contains additional citation context not shown here]

Radcliffe, 1992c. Nicholas J. Radcliffe. Genetic set recombination and its application to neural network topology optimisation. Neural Computing and Applications, 1(1), 1992.

.... operators respect schemata (and strictly transmit genes) Practical examples in which the principles are incompatible using natural formae include the travelling sales rep problem (Radcliffe, 1991b, where the formae are based on directed or undirected edges) neural network topology optimisation (Radcliffe, 1993, where they are based on nodes) and fixed size set and multiset problems (Radcliffe, 1992a, where they are based on set membership) Thus if a respectful operator is used, the non separability immediately means that assortment must fail to be achieved. The significance of this is that even when ....

Radcliffe, 1993. Nicholas J. Radcliffe. Genetic set recombination and its application to neural network topology optimisation. Neural Computing and Applications, 1(1):67--90, 1993.

....i.e. given some universal set, find the best subset of it according to some criterion. Examples include stock market portfolio optimisation (Shapcott, 1992) choosing k sites from n possible sites for retail dealers (George, 1994) and optimising the connectivity of a three layer neural network (Radcliffe, 1993). If the size of the subset is not fixed, the natural way to tackle this problem is by using a binary string the length of the universal set, using a 1 to indicate that the given element is in the subset. If, however, the size is fixed this is more problematical, because this constrains the number ....

Radcliffe, 1993. Nicholas J. Radcliffe. Genetic set recombination and its application to neural network topology optimisation. Neural Computing and Applications, 1(1):67--90, 1993.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC