D. Whitely, S. Dominic and D. Rajarshi. Genetic reinforcement learning with multilayer neural networks. Proceedings of the fourth international conference on genetic algorithms. Morgan Kaufmann 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....given that we have (arbitrarily) set the number of hidden layers it might be worthwhile to use GAs to determine the number of nodes in each layer. There have been several papers using GAs to design the overall topology of a network [Dodd 1989] Miller 1989] Whitley 1989] Whitley 1990] [Whitley 1991], Dodd 1991] and these have yeilded interesting results on the optimisation of network design. They have not, however, led to qualitatively new kinds of connectionist processes. Moreover, functionally equivalent networks with different topologies can easily fill up the population with networks ....

D. Whitely, S. Dominic and D. Rajarshi. Genetic reinforcement learning with multilayer neural networks. Proceedings of the fourth international conference on genetic algorithms. Morgan Kaufmann 1991.

....same form throughout the different stages of incremental evolution, making it simple and convenient to track progress. The neuron chromosomes are concatenations of the real valued weights on the input and output connections of the neuron. As is usual in ESP, burst mutation through delta coding [27] on these weights is used as needed to avoid premature convergence. If progress in evolution stagnates because the populations have converged, the populations are re initiated according to a Cauchy distribution around the current best solution. Burst mutation typically takes place at task ....

Whitley, D., Dominic, S., and Das, R. (1991). Genetic reinforcement learning with multilayer neural networks. In Belew, R. K., and Booker, L. B., editors, Proceedings of the Fourth International Conference on Genetic Algorithms, 562--569. San Francisco, CA: Morgan Kaufmann.

....of optimizing any classification structure or set of structures. In fact, people have tried optimizing most traditional machine learning structures as well as some nontraditiona l structures using GAs. These structures have ranged from neural network weights and topologies (Gruau Whitley, 1993; Whitley et al. 1990, 1991, 1993; Whitley Schaffer, 1992) to LISP programs (Koza, 1992) to regions of the instance space similar to decision trees induced by a splitting algorithm (Rendell, 1983, 1985; Sikora Shaw, 1994) to expertsystem rules (Montana, 1990) to weights for a game s evaluation function (Rendell, ....

Whitley, D., S. Dominic, and R. Das. 1991. Genetic reinforcemen t learning with multilayer neural networks. In Proceedings of the fourth international conference on genetic algorithms, 562569.

....Louis E. 26] Cortez, P. 325] Cremers, A. B. 610] Dachwald, Bernd, 93] Dagli, C. H. 184] Dagli, C. 655] Authors 17 Dai, Guiliang, 252] da Rocha Costa, Antonio Carlos, 226] Damour, S. 912] Danaher, S. 448] Danaher, Sean, 254] DaPonte, J. S. 538] Das, Rajarshi, [132, 955, 958, 959] Dasgupta, Dipankar, 299, 829, 830, 831, 832, 833] Datta, Amlan, 28] Davis, Lawrence, 656, 657] Dazhong, Wang, 560] De, Susmita, 213] De Angelo, S. 578] Deb, Kalyanmoy, 28] DeBaerdemaeker, J. 534] Deboeck, Guido, 658, 659] Deboeck, Tony, 658] Decarvalho, L. A. V. 858] ....

....G. 330] Dhawan, Atam P. 805, 867, 868] Dhawan, Atam, 62] Diessel, O. F. 869] Dill, Franz A. 660] Di Stefano, G. 29] Distefano, G. 592] Dobbins, R. W. 661] Dobnikar, Andrej, 30, 165, 246, 662] Dodd, Nigel, 663, 664, 806] Dolan, Charles P. 642] Dominic, Stephen, [132, 955, 956, 959] Dorado, Julian, 377] Dorey, Robert E. 26] Drabe, T. 440] Dracopoulos, Dimitris C. 31, 107] Dreiseitl, S. 441] Dreiseitl, Stephan, 166] Dress, W. B. 665, 666] Dube, D. 467] Dumitrescu, D. 331] Dumortier, F. 123] Duro, R. J. 32] Duro, Richard J. 82] Dybowski, R. ....

[Article contains additional citation context not shown here]

Darrell Whitley, Stephen Dominic, and Rajarshi Das. Genetic reinforcement learning with multilayer neural networks. In Belew and Booker [1034], pages 562--569. ga:Whitley91d.

....the explicit encoding on the chromosome of all the weights for a fixed number of neurons. Research has investigated performance on Boolean problems (e.g. Whitley and Hanson [78] Santos and Duro [65] Heistermann [27] Srinivas and Patnaik [72] and Maniezzo [47] polebalancing (Whitley et al. [76]) temporal pattern learning (e.g. Fukuda et al. 19] and classification (e.g. Korning [37] and Montana and Davis [52] The utility of using the GA over BP for training feedforward networks is unclear. Korning [37] obtained results only slightly worse than the best BP result obtained. Montana ....

Darrell Whitley, Stephen Dominic, and Rajarshi Das. Genetic reinforcement learning with multilayer neural networks. In Richard K. Belew and Lashon B. Booker, editors, Proceedings of the Fourth International Conference on Genetic Algorithms, pages 562--570, University of California, San Diego, 14--17 July 1991. Morgan Kaufmann.

....using the Euler s method (i.e. t 1) t) t) with a time step of = 0.02 second. The system is considered to be out of balance when the pendulum falls beyond 12 degrees from the vertical position or the cart runs beyond Sigma2:4 meters from the center. Previous approaches [2] 22] [33] to tackling the inverted pendulum problem employed a feedforward neural network using h; h; and as inputs, and the output was interpreted as the force applied to the cart. While the trained networks are able to balance the pendulum, four input variables are required to represent the system ....

D. Whitley, S. Dominic, and R. Das. Genetic reinforcement learning with multilayer neural networks. In Proceedings of the Fourth International Conference on Genetic Algorithms, pages 562--569, 1991.

....[63] makes a similar distinction and describes a GA which uses a set of independent subpopulations and structures the inter population communication with a ladder structure. These subpopulations are typically small, so they perform a localized search of the function. For example, Whitely [102] illustrates how a small 24 population can perform a locallized search in the context of neural network optimization problems. Inter population communication enables populations to combine disparate solutions and enables them to perform a global search. II.C.3 GAs with Local Search GA hybrids ....

Darrell Whitley, Stephen Dominic, and Rajarshi Das. Genetic reinforcement learning with multilayer neural networks. In Richard K. Belew and Lashon B. Booker, editors, Proceedings of the Fourth Intl. Conf. on Genetic Algorithms, pages 562--569. Morgan-Kaufmann, 1991.

....Morris [2] employing a (23 12 1010) topology, a (3 20 3) topology and a (120 60 20 21) topology, respectively. So at least the last one of them is much more complex than problems considered in publications on comparable topics ( 3] 4] 5] 6] 7] 8] 9] 10] 11] 12] 13] 14] [15]) Arguing with Miller et.al. 9] the search space of possible network topologies is infinitely large, not differentiable, complex, noisy, deceptive and multimodal. These attributes make random, enumerative, gradient descent or heuristic search methods unpracticable. Genetic algorithms, however, ....

....represented network. So the length of the genstring is equivalent to the number of potential connections allowed by the represented architecture. In weak representation schemes the genes correspond to more abstract network properties. Examples for such weak encodings can be found in [6] 7] or [15]. We agree with Miller et.al. 9] that weakschemes may be useful for capturing the architectural regularities of large networks rather efficiently . But their application also requires a much more detailed knowledge about both genetic and neural mechanisms. For this reason in our work we ....

[Article contains additional citation context not shown here]

D. Whitley, S. Dominic, R. Das, Genetic reinforcement learning with multilayer neural networks, in: Proc. Fourth Internat. Conf. Genetic Algorithms (Morgan Kaufmann, San Mateo, CA, 1991)

....GA which uses variable lengths of genotypes. This approach is well suited to the generation of highly recurrent neural networks. Another example of designing recurrent nets can be found in [41] Some reinforcement learning methods have also employed evolutionary methods with ANNs. For example in [49] a larger multi layer network design is evolved and in [1] interaction between learning as the adaptation of individual, and evolution as the adaptation of population is observed. In fact, there are quite a number of researchers concentrating on the relationship between evolution and learning ....

L.D. Whitley, S. Dominic, and R.Das. Genetic reinforcement learning with multilayer neural networks. In Belew and Booker, editors, Proceedings of Fourth International Conference on Genetic Algorithms, 1991.

....outperforms GA as well as NN in finding as satisfying solution. In the very long run, however, the NN alone is more precise [Kitano, 1990b] Long Genomes Create Problems It is a largely observed fact that GANN systems have difficulties with large networks, see for instance [Kitano, 1990b] [Whitley, 1991] and [White, 1993] While functions like XOR have been successfully applied to nearly all approaches, reports of complex tasks with many input and output nodes are rather rare. Structural Functional Mapping Problem One problem of encoding neural network information into a genome is what Whitley ....

: D. Whitley, S. Dominic, R. Das: "Genetic Reinforcement Learning with Multilayer Neural Networks", in: Proceedings of the Fourth International Conference on Genetic Algorithms, pp. 562-569, Morgan Kaufmann.

..... 25 4.2 Future Work . 26 Abstract In the Neural network genetic algorithm community, rather limited success in the training of neural networks by genetic algorithms has been reported. In a paper by Whitley (1991), he claims that, due to the multiple representations problem , genetic algorithms will not e#ectively be able to train multilayer perceptrons, whoes chromosomal representation of its weights exceeds 300 bit s. In the following paper, by use of a real life problem , known to be non trivial, and ....

....are usually relatively short. Montana uses strings consisting of eleven real numbers, which must be assumed to be insu#cient for setting the building block hypothesis to work. Caudell Dolan use 288 bits; not much either. The above conditions are summarised in a paper by Darrell Whitley [Whitley et al..91] in which it is concluded, that GAs are unsuited for the training of perceptrons, in cases where the coding of the weights consists of more than 300 bits. For this aim he proposes the use of a hybrid, a genetic hillclimber , called GENITOR, which as the name implies is primarily based on ....

Whitley, D., Dominic, S. and Das, R.(1991). Genetic Reinforcement Learning with Multilayer Neural Networks. Proceedings of the Fourth International Conference on Genetic Algorithms, pp. 562-569. Morgan Kaufmann.

.... have been used to tune the parameters of other training schemes, including initial weight configurations (Belew, McInerny Shraudolph [1] All of these approaches have associated problems, which have been discussed by Montana Davis [16] Radcliffe [20, 21] Belew et al. 1] and Whitley et al. [37]. Principal among these, and appearing in many different guises, is a permutational redundancy associated with the arbitrariness of labels of topologically equivalent hidden nodes. Specifically, to take an extreme case, in a fully connected feed forward, layered network with a single hidden layer ....

....labelling of hidden units, the search space is enormously enlarged. While optima usually become more numerous by a comparable factor, the global nature of genetic search tends to make navigation through the enlarged search space very difficult (Radcliffe [20, 21] Belew et al. 1] Whitley et al. [37] and section 3.2) Genetic algorithms are sensitive to the potential for redundant representations in a way that most other search schemes (for example, gradient techniques and 1 albeit one which often has a rather poorly defined objective function h 1 h 2 h 3 i 1 i 2 i 3 i 4 i 5 o 1 o 2 o 3 o ....

Darrell Whitley, Stephen Dominic, and Rajarshi Das. Genetic Reinforcement Learning With Multilayer Neural Networks. In Proceedings of the Fourth International Conference on Genetic Algorithms. Morgan Kaufmann (San Mateo), 1991.

....and nondifferentiable space. It does not depend on the gradient information of the error (or fitness) function, thus is particularly appealing when the gradient information is unavailable or very costly to get. For example, the evolutionary approach has been used in reinforcement learning [47, 64, 69, 70, 71], recurrent network learning [45, 64, 72] and higher order network learning [56, 57] Moreover, the same evolutionary algorithm can be used to training many different networks regardless of whether they are feedforward networks, recurrent networks, or higher order networks. The general ....

D. Whitley, S. Dominic, and R. Das. Genetic reinforcement learning with multilayer neural networks. In R. K. Belew and L. B. Booker, editors, Proc. of the Fourth Int'l Conf. on Genetic Algorithms, pages 562--570. Morgan Kaufmann, San Mateo, CA, 1991.

....(i.e. 40 minutes of simulated time) or the allowed number of iterations are used up. The problem has been solved by the researchers of both neural networks and genetic algorithms. The combination of the two are also used, for example, a fixed network is trained with genetic reinforcement learning [9] and other method used a genetic cascade learning for sequentially building the net to perform the task. Our method is an alternative neurogenetic approach where network architecture is also evolved in an implicitly parallel fashion. 4 Evolving Neuro Controllers with sGA Figure 1 shows the ....

....undergoes a hypermutation on its high level. Also the feasible individuals which have fewer nodes and links get higher reproductive chance relative to the competing feasible individuals with more complex structures. Each feasible net (individual) is trained through genetic reinforcement learning [9], where the learning process is also an object of evolution. The learning process starts with providing the initial state of the system to the net and the net s output response is applied to the simulated system. The output of the net is either 0.0 (push left) or 1.0 (push right) representing the ....

[Article contains additional citation context not shown here]

D. Whitley, S. Dominic, and R. Das. Genetic reinforcement learning with multilayer neural networks. In

....and nondifferentiable space. It does not depend on the gradient information of the error (or fitness) function, thus is particularly appealing when the gradient information is unavailable or very costly to get. For example, the evolutionary approach has been used in reinforcement learning [46, 62, 66, 67], recurrent network learning [44, 62, 68] and higher order network learning [55] It should be noted that the same evolutionary algorithm can be used to training many different networks, X. Yao: Evolutionary Artificial Neural Networks 8 such as feedforward networks, recurrent networks, higher ....

D. Whitley, S. Dominic, and R. Das. Genetic reinforcement learning with multilayer neural networks. In R. K. Belew and L. B. Booker, editors, Proc. of the Fourth Int'l Conf. on Genetic Algorithms, pages 562--570. Morgan Kaufmann, San Mateo, CA, 1991.

.... learning: GAs have been used for many machine learning applications, including classification and prediction tasks such as the prediction of dynamical systems [75] weather prediction [92] and prediction of protein structure (e.g. 95] GAs have also been used to design neural networks (e.g. [15, 25, 47, 48, 67, 77, 81, 94, 105]) to evolve rules for learning classifier systems (e.g. 54, 57] or symbolic production systems (e.g. 46] and to design and control robots (e.g. 29, 31, 50] For an overview of GAs in machine learning, see [64, 65] ffl Economic models: GAs have been used to model processes of ....

L. D. Whitley, S. Dominic, and R. Das. Genetic reinforcement learning with multilayer neural networks. In R. K. Belew and L. B. Booker, editors, Proceedings of the Fourth International Conference on Genetic Algorithms, pages 562--569, San Mateo, CA, 1991. Morgan Kaufmann.

....(e.g. Belew et al. (1990) there have been few attempts to solve it. Indeed, a number of workers have simply removed crossover from the algorithm, e.g. Bornholdt Graudenz, 1992; de Garis, 1990; Nolfi et al. 1990) which seems drastic given the centrality of recombination to the GA 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 ....

Whitley, D., Dominic, S., & Das, R. 1991. Genetic reinforcement learning with multilayer neural networks. Pages 562--569 of: Belew, R.K., & Booker, LB. (eds), Proceedings of the fourth international conference on Genetic Algorithms. Morgan Kaufmann.

....if a population of highly fit individuals evolves as a result of iterating this procedure. The GA has been used in many machine learning contexts, such as evolving classification rules (e.g. Packard, 1990; De Jong and Spears, 1991) evolving neural networks (e.g. Miller, Todd, Hegde, 1989; Whitley, Dominic, Das, 1991), classifier systems (e.g. Holland, 1986; Smith, 1980) and automatic programming (e.g. Koza, 1990) In many of these cases there is no closed form fitness function ; the evaluation of each individual (or collection of individuals) is obtained by running it on the particular task being ....

.... factors that contribute to the difficulty of search for a GA, and (3) concluding that any successful research effort into the theory of GA performance must take into account this multiplicity rather than concentrating on only one factor (e.g. deception; see Goldberg, 1989b; Liepins Vose, 1990; Whitley, 1991; and Das Whitley, 1991) While this paper deals with learning real valued functions on bit strings, the discussion and results presented here are relevant to researchers using GAs in other contexts as well, for a number of reasons. First, it is important for researchers using a particular ....

[Article contains additional citation context not shown here]

Whitley, L. D., Dominic, S., and Das, R. (1991). Genetic reinforcement learning with multilayer neural networks. In R. K. Belew and L. B. Booker (Eds.), Proceedings of the Fourth International Conference on Genetic Algorithms, 562--569. San Mateo, CA: Morgan Kaufmann.

No context found.

: D. Whitley, S. Dominic, R. Das: "Genetic Reinforcement Learning with Multilayer Neural Networks", in: Proceedings of the Fourth International Conference on Genetic Algorithms, pp. 562-569, Morgan Kaufmann.

No context found.

D. Whitley, S. Dominic, and R. Das. Genetic reinforcement learning with multilayer neural networks. In Proceedings of the Fourth International Conference on Genetic Algorithms, pages 562--569. Morgan Kaufmann, 1991.

No context found.

Darrell Whitley, Stephen Dominic, and Rajarshi Das. Genetic Reinforcement Learning with Multilayer Neural Networks. In Proceedings of the International Conference on Genetic Algorithms, pages 562--5769, 1991.

No context found.

D. Whitley, S. Dominic, and R. Das. Genetic Reinforcement Learning with Multilayer Neural Networks. In R. K. Belew and L. B. Booker, editors, Proceedings of the Fourth International Conference on Genetic Algorithms, pages 562--569. Morgan Kaufmann, San Mateo, CA, 1991.

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