Neuroevolution with Analog Genetic Encoding

Dürr, Peter; Mattiussi, Claudio; Floreano, Dario

doi:10.1007/11844297_68

Cited by 52 publications

(39 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Some authors use binary or real encoding (representation of the networks in a binary or real number string) [37], [38], or indirect coding [39], [40], but G-Prop evolves the initial parameters of the network (initial weights and learning constants) using specific genetic operators. At the lowest level, an MLP is an object instantiated from the MLP C++ class.…”

Section: The Methodsmentioning

confidence: 99%

Evolutionary system for prediction and optimization of hardware architecture performance

Castillo

Merelo

Moretó

et al. 2008

2008 IEEE Congress on Evolutionary Computation (IEEE World Congress on Computational Intelligence)

View full text Add to dashboard Cite

Abstract-The design of computer architectures is a very complex problem. The multiple parameters make the number of possible combinations extremely high. Many researchers have used simulation, although it is a slow solution since evaluating a single point of the search space can take hours. In this work we propose using evolutionary multilayer perceptron (MLP) to compute the performance of an architecture parameter settings. Instead of exploring the search space, simulating many configurations, our method randomly selects some architecture configurations; those are simulated to obtain their performance, and then an artificial neural network is trained to predict the remaining configurations performance. Results obtained show a high accuracy of the estimations using a simple method to select the configurations we have to simulate to optimize the MLP. In order to explore the search space, we have designed a genetic algorithm that uses the MLP as fitness function to find the niche where the best architecture configurations (those with higher performance) are located. Our models need only a small fraction of the design space, obtaining small errors and reducing required simulation by two orders of magnitude.

show abstract

Section: The Methodsmentioning

confidence: 99%

Evolutionary system for prediction and optimization of hardware architecture performance

Castillo

Merelo

Moretó

et al. 2008

2008 IEEE Congress on Evolutionary Computation (IEEE World Congress on Computational Intelligence)

View full text Add to dashboard Cite

show abstract

“…Other related task decomposition approaches in the literature include analog genetic encoding (AGE) [10,11,36,37]. Nevertheless, unlike RMGEP, AGE uses alphabetic labels or tokens which separate coding from non-coding parts of the genome to evolve parts of a system.…”

Section: Regulatory Multigenic Gene Expression Programmingmentioning

confidence: 99%

Evolving robot sub-behaviour modules using Gene Expression Programming

Mwaura

Keedwell

2014

Genet Program Evolvable Mach

View full text Add to dashboard Cite

Many approaches to AI in robotics use a multi-layered approach to determine levels of behaviour from basic operations to goal-directed behaviour, the most well-known of which is the subsumption architecture. In this paper, the performances of the unigenic gene expression programming (ugGEP) and multigenic GEP (mgGEP) in evolving robot controllers for a wall following robot is analysed. Additionally, the paper introduces Regulatory Multigenic Gene Expression Programming (RMGEP), a new evolutionary technique that can be utilised to automatically evolve modularity in robot behaviour. The proposed technique extends the mgGEP algorithm, by incorporating a regulatory gene as part of the GEP chromosome. The regulatory gene, just as in systems biology, determines which of the genes in the chromosome to express and therefore how the controller solves the problem. In the initial experiments, the proposed algorithm is implemented for a robot wall following problem and the results compared to that of ugGEP and mgGEP. In addition to the wall following behaviour, a robot foraging behaviour is implemented with the aim of investigating whether the position of a specific module (sub-expression tree (ET)) in the overall ET is of importance when coding for a problem.

show abstract

“…The initial population contained genomes of random length, , ranging from 2 to 20 genes, with indices chosen at random from [1,100], and values chosen at random from [−30, 30]. With this setup, no one genome contains all of the 100 available frequencies, but with very high probability all frequencies are present in the population.…”

Section: Setupmentioning

confidence: 99%

“…Indirect or generative encoding schemes for neural network phenotypes [1,5,6,9] offer the potential of allowing very large networks to be represented compactly. In previous work [7], we showed that encoding neural network weight matrices indirectly as a set of Fourier-type coefficients can reduce the search space dimensionality and help to discover more 'regular' networks which are simpler in the Kolmogorov sense (the program required to encode them is much shorter).…”

Section: Introductionmentioning

confidence: 99%

Generalized compressed network search

Srivastava

Schmidhuber

Gomez

2012

Proceedings of the 14th Annual Conference Companion on Genetic and Evolutionary Computation

View full text Add to dashboard Cite

Abstract. This paper presents initial results of Generalized Compressed Network Search (GCNS), a method for automatically identifying the important frequencies for neural networks encoded as Fourier-type coefficients (i.e. "compressed" networks [7]). GCNS is a general search procedure in this coefficient space -both the number of frequencies and their value are automatically determined by employing the use of variable-length chromosomes, inspired by messy genetic algorithms. The method achieves better compression than our previous approach, and promises improved generalization for evolved controllers. Results for a high-dimensional Octopus arm control problem show that a high fitness 3680-weight network can be encoded using less than 10 coefficients using the frequencies identified by GCNS.

show abstract

Neuroevolution with Analog Genetic Encoding

Cited by 52 publications

References 17 publications

Evolutionary system for prediction and optimization of hardware architecture performance

Evolutionary system for prediction and optimization of hardware architecture performance

Evolving robot sub-behaviour modules using Gene Expression Programming

Generalized compressed network search

Contact Info

Product

Resources

About