Moriarty D E and Miikkulainen R (1997),. Forming Neural Networks through Efficient and Adaptive Coevolution. Evolutionary Computation Volume 5, pp. 373-399.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....scheme where it assumes architecture consists of various segments or areas. Each segment or area will define a set of neurons, their spatial arrangement and their efferent connectivity. Several high level coding schemes like graph generation system [49] Symbiotic Adaptive Neuro Evolution (SANE) [62] [65] marker based genetic coding [36] L systems [13] cellular encoding [38] fractal representation [58] etc are some of the rugged techniques. 1 5 3 4 From To 1 2 3 4 5 Bias Gene 1 0 0 0 0 0 0 000000 2 0 0 0 0 0 0 000000 3 1 1 0 0 0 1 110001 4 1 1 0 0 0 1 110001 5 0 0 1 1 0 1 ....

Moriarty D E and Miikkulainen R (1997),. Forming Neural Networks through Efficient and Adaptive Coevolution. Evolutionary Computation Volume 5, pp. 373-399.

....the network and in fact may lead to better scores in the final output. This is the main reason why competition is not the only factor in network optimization, and co operative behavior is desired. Various mechanisms have been 14 suggested for implementing this with varying levels of complexity [40]. The Optimization Cycle approach that we used is most similar to greedy strategy suggested in [37] In the Optimization Cycle, nodes receive reward based on the final network output. There are 9 stages to the optimization cycle, 1 for each node in the network. The network is configured with the ....

Moriarty, D.E. and R. Miikkulaiinen, Forming Neural Networks through Efficient and Adaptive Coevolution. Evolutionary Computation, 1998. 5(4).

....to produce neural networks in general (see [12] for an early review) In contrast to that work, an LCS based approach is co evolutionary, the aim being to develop a number of (small) cooperative neural networks to solve the given task, as opposed to the evolution of one (large) network. SANE [11] is most similar to the work described here, however SANE coevolves individual neurons rather than small networks of neurons as rules. In this paper we implement the new rule representation within XCS, using the specification given in [4] Results are presented for the single step multiplexer, a ....

Moriarty, D.E. & Miikulainen, R. (1997) Forming Neural Networks Through Efficient and Adaptive Coevolution. Evolutionary Computation 5(2): 373-399.

....weights to the input layer (environmental states) and output layer (control actions) In the genetic algorithm framework, effective neurocontrollers are allowed to produce offspring, which promotes the propagation of effective neurons (genetic material) in the population. Moriarty and Miikkulainen [7] have developed a state of the art genetic algorithm, Symbiotic Adaptive Neuro Evolution (SANE) that uses implicit fitness sharing to ensure genetic diversity in the controller population, while allowing for an aggressive search in the solution space. Implicit fitness sharing entails the search ....

....diversity makes convergence to a local optimum far less likely than in a standard genetic algorithm implementation. A robust search for the global optimum is thus ensured. Also, several parallel searches for partial solutions should prove more effective than a single search for the entire solution [7]. Process u d y p r y p z 1 Fig. 2. SANE neurocontroller closed loop architecture (r reference point; y P present process variables; d load disturbances; z 1 time delay; u manipulated variables) In the SANE framework, the neurocontroller is implemented in the closed ....

D.E. Moriarty and R. Miikkulainen, 1998. Forming Neural Networks through Efficient and Adaptive Coevolution. Evolutionary Computation, 5(4), 373-399.

....decomposition. 4 Evolutionary Computation Volume 8, Number 1 Cooperative Coevolution The application of EAs to the construction of artificial neural networks has also motivated extensions to the basic evolutionary model for the support of coadapted subcomponents. For example, in the SANE system (Moriarty, 1997; Moriarty and Miikkulainen, 1997) each individual in the population represents a single neuron by specifying which input and output units it connects to and the weights on each of its connections. A collection of neurons selected from the population constitutes a specification for constructing a ....

....4 Evolutionary Computation Volume 8, Number 1 Cooperative Coevolution The application of EAs to the construction of artificial neural networks has also motivated extensions to the basic evolutionary model for the support of coadapted subcomponents. For example, in the SANE system (Moriarty, 1997; Moriarty and Miikkulainen, 1997), each individual in the population represents a single neuron by specifying which input and output units it connects to and the weights on each of its connections. A collection of neurons selected from the population constitutes a specification for constructing a complete neural network. A ....

Moriarty, D. E. and Miikkulainen, R. (1997). Forming neural networks through efficient and adaptive coevolution. Evolutionary Computation, 5(4):373--399.

....that have to be solved to learn the problem. This approach allows a way to impose more structure on the learning process and by this to accelerate it. A second approach which is investigated in parallel is the use of a rule evolution mechanism inspired by different paradigms of neuro evolution [2, 5, 6]. In these approaches, unlike in many other neuro evolutionary approaches, not a population of neural networks is evolved. Instead the evolution takes place on the level of a population of substructures, e.g. of single neurons. We transfer this idea from neurons to rules, i.e. we perform an ....

....Thomas Uthmann Any rule, as above, consists of conditions triggering it and of actions and or rules following it. Any condition and action is fine tuned by certain parameters and the particular assembly of conditions and actions constitutes a given rule. Inspired by the SANE and EuSANE approach [5, 6], we coevolve two things: one the one hand we optimize the parameters determining the condition and actions, on the other hand we evolve their arrangement into a given rule. This arrangement is also called blueprint, following the SANE nomenclature. The fitness is given by solving given tasks and, ....

D. Moriarty and R. Miikkulainen. Forming neural networks through efficient and adaptive co-evolution. Evolutionary Computation, 5:373--399, 1997.

....(control actions) In an EA framework, effective neurocontrollers produce offspring, which propagates effective neurons (genetic material) in the population. This genetic propagation of effective neuron structures is key to solving the combinatorial nature of neurocontroller parameter estimation [4]. B. Memetic algorithms EA s propagate effective neuron structures by varying the sample distribution in the solution space, depending upon the evaluation of the objective (fitness) function. This selection biases the search towards regions of the solution space where near optimal solutions have ....

....II) is employed to ensure global reliability, while performing an aggressive explorative search. Particle Swarm Optimisation (PSO) a cultural evolution method, is used for local exploitative search after each EA generation. 1) Symbiotic evolutionary algorithm Similar to the SANE algorithm [4], the symbiotic EA (dashed box in figure II) maintains both a neuron and a network population. Each member of the neuron population encodes a hidden neuron, with weights from the input layer to the output layer. While SANE maintains a single neuron population, SMNE s neuron population is comprised ....

[Article contains additional citation context not shown here]

D.E. Moriarty and R. Miikkulainen, "Forming Neural Networks through Efficient and Adaptive Coevolution", Evolutionary Computation, 5(4), pp. 373-399, 1998.

....(control actions) In an EA framework, effective neurocontrollers produce offspring, which propagates effective neurons (genetic material) in the population. This genetic propagation of effective neuron structures is key to solving the combinatorial nature ofneurocontroller parameter estimation [4]. B. Memetic algorithms EA s propagate effective neuron structures by varying the sample distribution in the solution space, depending upon the evaluation of the objective (fitness) function. This selection biases the search towards regions of the solution space where near optimal solutions have ....

....II) is employed to ensure global reliability, while performing an aggressive explorative search. Particle Swarm Optimisation (PSO) a cultural evolution method, is used for local exploitative search after each EA generation. 1) Symbiotic evolutionary algorithm Similar to the SANE algorithm [4], the symbiotic EA (dashed box in figure II) maintains both a neuron and a network population. Each member of the neuron population encodes a hidden neuron, with weights from the input layer to the output layer. While SANE maintains a single neuron population, SMNE s neuron population is comprised ....

[Article contains additional citation context not shown here]

D.E. Moriarty and R. Miikkulainen, "Forming Neural Networks through Efficient and Adaptive Coevolution", Evolutionary Computation, 5(4), pp. 373-399, 1998.

....used. To make sure that the absolute quality of the hosts is increasing, they will not only be tested against a subset of parasites, but also against the best hosts of previous generations (i.e. a hall of fame, Rosin, 1997) As the evolutionary technique we will use the SANE neuro evolution method (Moriarty and Miikkulainen, 1997), because it has been shown effective in the go domain (Richards et al. 1998) The results show that the learning speed is increased by using the co evolutionary techniques and the level of play is not limited by existing opponents. 2 THE GAME OF GO The game of go is an ancient board game which ....

....combination with the Japanese scoring we use: it is possible to reduce the score of the opponent by letting him fill the empty intersections by just throwing in stones in his territory. However, on such small boards this will not be a problem. 3. 2 SANE SANE (Symbiotic Adaptive Neuro Evolution; Moriarty and Miikkulainen, 1997; Richards et al. 1998) differs from standard evolutionary algorithms in that instead of evolving complete neural networks, a population of neurons and a population of blueprints (that specify which neurons to combine into a neural network) are evolved. By evolving neurons instead of complete ....

[Article contains additional citation context not shown here]

Moriarty, D. and Miikkulainen, R. (1997). Forming neural networks through efficient and adaptive coevolution. Evolutionary Computation, 5:373--399.

....neural network. predators face is coordinating the chase so that in the end the prey has nowhere to go, which requires a more long term strategy. 4 The Multi Agent ESP Approach The Enforced Subpopulations Method (The ESP 1 ; 7, 8] is an extension of Symbiotic, Adaptive NeuroEvolution (SANE; [17, 18, 16]) SANE is a method of neuro evolution that evolves a population of neurons instead of complete networks. Neurons are selected from the population to form the hidden layer of a neural network, which is evaluated on the problem. The fitness is then passed back to all the partaking neurons of the ....

Moriarty, D. E., and Miikkulainen, R. (1997). Forming neural networks through efficient and adaptive co-evolution. Evolutionary Computation, 5:373--399.

....over generations, and strengthening the analogy with biological evolution. 1 Introduction Neuroevolution (NE) the artificial evolution of neural networks using genetic algorithms, has shown great promise in complex reinforcement learning tasks (Gomez and Miikkulainen 1999; Gruau et al. 1996; Moriarty and Miikkulainen 1997; Potter et al. 1995; Whitley et al. 1993) Neuroevolution searches through the space of behaviors for a network that performs well at a given task. This approach to solving complex control problems represents an alternative to statistical techniques that attempt to estimate the utility of ....

....et al. 1996) NE is a promising approach to solving reinforcement learning problems for several reasons. Past studies have shown NE to be faster and more efficient than reinforcement learning methods such as Adaptive Heuristic Critic and Q Learning on single pole balancing and robot arm control (Moriarty 1997; Moriarty and Miikkulainen 1996) Because NE searches for a behavior, it is effective in problems with large state spaces. In addition, memory is easily represented through recurrent connections in neural networks, making the method a natural choice for learning non Markovian tasks (Gomez and ....

Moriarty, D. E., and Miikkulainen, R. (1997). Forming neural networks through efficient and adaptive co-evolution. Evolutionary Computation, 5:373--399.

....Policies Distributed EARL systems using neural net representations have also been designed. In (Potter De Jong, 1995) separate populations of neurons evolve, with the evaluation of each neuron based on the fitness of a collaboration of neurons selected from each population. In SANE (Moriarty Miikkulainen, 1996a, 1998), two separate populations are maintained and evolved: a population of neurons and a population of network blueprints. The motivation for SANE comes from our a priori knowledge that individual neurons are fundamental building blocks in neural networks. SANE explicitly decomposes the neural network ....

....solutions are built by combining individuals. Because no individual can solve the task on its own, the evolutionary algorithm will search for several complementary individuals that together can solve the task. Evolutionary pressures are therefore present to prevent convergence of the population. Moriarty and Miikkulainen (1998) showed how the inherent diversity and specialization in SANE allow it to adapt much more quickly to changes in the environment than standard, convergent evolutionary algorithms. Finally, if the learning system can detect changes in the environment, even more direct response is possible. In the ....

[Article contains additional citation context not shown here]

Moriarty, D. E., & Miikkulainen, R. (1998). Forming neural networks through efficient and adaptive co-evolution. Evolutionary Computation, 5 (4), 373--399.

No context found.

Moriarty, D.E. and Miikkulainen, R. (1998). Forming Neural Networks through Efficient and Adaptive Coevolution.

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