## Delta Coding: An Iterative Search Strategy for Genetic Algorithms (1991)

Venue: | Proceedings of the Fourth International Conference on Genetic Algorithms |

Citations: | 31 - 1 self |

### BibTeX

@INPROCEEDINGS{Whitley91deltacoding:,

author = {D. Whitley and K. Mathias and P. Fitzhorn},

title = {Delta Coding: An Iterative Search Strategy for Genetic Algorithms},

booktitle = {Proceedings of the Fourth International Conference on Genetic Algorithms},

year = {1991},

pages = {77--84},

publisher = {Morgan Kaufmann}

}

### Years of Citing Articles

### OpenURL

### Abstract

A new search strategy for genetic algorithms is introduced which allows iterative searches with complete reinitialization of the population preserving the progress already made toward solving an optimization task. Delta coding is a simple search strategy based on the idea that the encoding used by a genetic algorithm can express a distance away from some previous partial solution. Delta values are added to a partial solution before evaluating the fitness; the delta encoding forms a new hypercube of equal or smaller size that is constructed around the most recent partial solution.

### Citations

288 |
The CHC Adaptive Search Algorithm: How to Have Safe Search When Engaging in Nontraditional Genetic Recombination
- Eshelman
- 1991
(Show Context)
Citation Context ...initially described this method of implementation for serial GAs [5]. The motivation is to maximize the number of schemata sampled during genetic search. Cataclysmic mutation, as described by Eshelman=-=[4]-=- also provides a method of restarting the GA search by reinitializing the population. During reinitialization, the best string in the population at the time of convergence is used as a template by ran... |

257 |
A Connectionist Machine for Genetic Hillclimbing
- Ackley
- 1987
(Show Context)
Citation Context ...rsity maintenance such as mutation, adaptive mutation and distributed genetic algorithms[12] as well as more divergent variations on genetic search such as Ackley's connectionist genetic hill-climber =-=[1]-=- and Shaefer's ARGOT Strategy [10]. Diversity maintenance is critical, since the search reaches a plateau and stops if the entire population has converged to a single solution (or is populated with st... |

160 |
Sizing Populations for Serial and Parallel Genetic Algorithms
- Goldberg
- 1989
(Show Context)
Citation Context ... the search is repeated. No mutation is used. This strategy has been shown to give good performance on certain test problems. Goldberg initially described this method of implementation for serial GAs =-=[5]-=-. The motivation is to maximize the number of schemata sampled during genetic search. Cataclysmic mutation, as described by Eshelman[4] also provides a method of restarting the GA search by reinitiali... |

135 |
Genetic algorithm and neural networks: Optimizing connexions and connectivity
- Whitley, Starkweather, et al.
- 1990
(Show Context)
Citation Context ...in between (e.g. 00111) would result in very poor solutions. Therefore, recombination between the dissimilar parents would generally produce worse solutions providing inconsistent hyperplane feedback =-=[13]-=-. Delta coding may help resolve such difficulties by redefining new hyperplane mappings in N-space and redefining the hypercube with respect to only one of the competing solutions. ACKNOWLEDGEMENTS Th... |

118 |
How Genetic Algorithms work: a critical look at implicit parallelism
- Grefenstette, Baker
- 1989
(Show Context)
Citation Context ...tions found by genetic algorithms were sometimes close to a precise solution such that we could complement a small number of bits and obtain a globally optimal solution. Also, recent theoretical work =-=[3]-=- shows that selective pressure on hyperplanes is not uniform during genetic search; bits that are selected against early on may actually be preferred as the search narrows to concentrate on specialize... |

100 | Fundamental Principles of Deception in Genetic Search, Foundations of Genetic Algorithms
- Whitley
- 1991
(Show Context)
Citation Context ...lta coding may help to solve deceptive problems. Deceptive problems exist because the relationship between specific hyperplanes is such that different hyperplane competitions have conflicting winners =-=[14]-=-; however, since delta codingschanges the way hyperspace is defined with respect to the objective function at each iteration, the same partitions of hyperspace will not exist from one delta iteration ... |

94 |
Micro-genetic algorithms for stationary and non-stationary function optimization
- Krishnakumar
- 1989
(Show Context)
Citation Context ...rete stages of search may in fact be advantageous; the other strategy we have looked at that uses discrete stages of search is micro-GAs. 2.4 MICRO-GAs AND OTHER REINITIALIZATION METHODS The micro-GA =-=[7]-=- uses a small population (e.g. five strings). The population is measured for convergence either by genotype convergence or phenotype convergence. If the population has converged the best string is kep... |

88 | Dynamic parameter encoding for genetic algorithms
- Schraudolph, Belew
- 1992
(Show Context)
Citation Context ...d to drive strategies within ARGOT such as adjusting parameter resolution and roving parameter boundary locations [10]. Dynamic parameter encoding (DPE) is a strategy explored by Schraudoph and Belew =-=[9]-=- which has many similarities to ARGOT; the accuracy of the encoded parameters are dynamically adjusted to increase the resolution of the answer and to zoom in on the most promising area of the search ... |

79 |
Optimizing neural networks using faster, more accurate genetic search
- Whitley, Hanson
- 1989
(Show Context)
Citation Context ...ction guarantees that the best solutions found so far will be held undisturbed in the population until a better solution is located. GENITOR has displayed superior performance in many problem domains =-=[15]-=-. The perpetuation of superior strings guides the search by providing more bias or probability of commitment toward better solutions thereby introducing a kind of hillclimbing to genetic search, with ... |

78 |
GENITOR II: A distributed genetic algorithm
- Whitley, Starkweather
- 1990
(Show Context)
Citation Context ...diversity is a pervasive problem with genetic algorithms. Several strategies have been developed to support diversity maintenance such as mutation, adaptive mutation and distributed genetic algorithms=-=[12]-=- as well as more divergent variations on genetic search such as Ackley's connectionist genetic hill-climber [1] and Shaefer's ARGOT Strategy [10]. Diversity maintenance is critical, since the search r... |

66 |
Optimization Using Distributed Genetic Algorithms
- STARKWEATHER, WHITLEY
- 1991
(Show Context)
Citation Context ...rect for sampling biases that may occur in individual populations. The GENITOR II algorithm not only shows improved performance in terms of more accurate solutions, it also requires less overall work =-=[11]-=-. 2.3 THE ARGOT STRATEGY AND DYNAMIC PARAMETER ENCODING The ARGOT strategy (Adaptive Representation Genetic Optimizer Technique) combines many ideas to improve optimization; the results appear to be s... |

55 |
Improving search in genetic algorithms
- Booker
- 1987
(Show Context)
Citation Context ...on between the points from one parent to pass on to the offspring and the information outside of the points from the other parent to pass on to the offspring. The reduced surrogate crossover operator =-=[2]-=- used in these experiments, can ensure that offspring are not duplicates of the parents. The goal of genetic algorithms is to exploit hyperplane information feedback by evaluating the relative fitness... |

2 |
Delta Coding Strategies for Genetic Algorithms
- Mathias
- 1991
(Show Context)
Citation Context ... the Euclidean distance between the corresponding target and transformed points, summing the squared distances for all point pairs. The goal was to minimize the error. (For implementation details see =-=[8]-=-). The 3-D to 2-D transformational problem was defined by a specific application and motivated the 2-D test problems [6]. The goal in the 3-D transformation problem was to provide a set of transformat... |

2 |
The ARGOT Strategy: Adaptive Representative Genetic Optimizer Technique." Genetic Algorithms and their Applications
- Shaefer
- 1987
(Show Context)
Citation Context ...n, adaptive mutation and distributed genetic algorithms[12] as well as more divergent variations on genetic search such as Ackley's connectionist genetic hill-climber [1] and Shaefer's ARGOT Strategy =-=[10]-=-. Diversity maintenance is critical, since the search reaches a plateau and stops if the entire population has converged to a single solution (or is populated with strings whose potential offspring ar... |

1 |
A Method for Near Optimal
- Nickerson, Fitzhorn, et al.
- 1991
(Show Context)
Citation Context ...oint pairs. The goal was to minimize the error. (For implementation details see [8]). The 3-D to 2-D transformational problem was defined by a specific application and motivated the 2-D test problems =-=[6]-=-. The goal in the 3-D transformation problem was to provide a set of transformation parameters that would map a set of 3-D points onto a target set of points in a 2-D domain. As in the 2-D transformat... |