## Evolutionary Learning Of Fuzzy Rules: Competition And Cooperation (1996)

Citations: | 59 - 8 self |

### BibTeX

@MISC{Bonarini96evolutionarylearning,

author = {Andrea Bonarini},

title = {Evolutionary Learning Of Fuzzy Rules: Competition And Cooperation},

year = {1996}

}

### Years of Citing Articles

### OpenURL

### Abstract

We discuss the problem of learning fuzzy rules using Evolutionary Learning techniques, such as Genetic Algorithms and Learning Classifier Systems. We present ELF, a system able to evolve a population of fuzzy rules to obtain a sub-optimal Fuzzy Logic Controller. ELF tackles some of the problems typical of the Evolutionary Learning approach: competition and cooperation between fuzzy rules, evolution of general fuzzy rules, imperfect reinforcement programs, fast evolution for real-time applications, dynamic evolution of the focus of the search. We also present some of the results obtained from the application of ELF to the development of Fuzzy Logic Controllers for autonomous agents and for the classical cart-pole problem. INTRODUCTION Genetic Algorithms (GAs)[13] and Learning Classifier Systems (LCS)[7][8] emerged in the last years as powerful Evolutionary Learning (EL) techniques to identify systems that optimize some cost function. The cost function provides a reinforcement that gui...

### Citations

2061 |
Genetic algorithms in search, optimization, and machine learning
- Goldberg
- 1989
(Show Context)
Citation Context ...e of the results obtained from the application of ELF to the development of Fuzzy Logic Controllers for autonomous agents and for the classical cart-pole problem. INTRODUCTION Genetic Algorithms (GAs)=-=[13]-=- and Learning Classifier Systems (LCS)[7][8] emerged in the last years as powerful Evolutionary Learning (EL) techniques to identify systems that optimize some cost function. The cost function provide... |

329 |
Neural Networks and Fuzzy Systems: A Dynamical Systems Approach to
- Kosko
- 1992
(Show Context)
Citation Context ...difficulty to evaluate the correctness of a reinforcement program in different experiments with autonomous agents [2][3][5], and other fuzzy control applications, such as: the cart-pole balancing [14]=-=[17]-=-, the spacecraft autonomous rendez-vous [15] and the truck backing-up [17]. We have found that it is common to have critical states that are not considered by reinforcement programs and that "naturall... |

301 | Classifier fitness based on accuracy
- Wilson
- 1995
(Show Context)
Citation Context ...erferes with the evolution of specialized rules. The solutions we have adopted to solve these problems can also be extended to similar problems arising with other EL Algorithms, in particular LCS [26]=-=[27]-=-. Cooperation and competition Usually, the action an FLC produces for the actuators does not depend on one rule only, but on different rules, each one triggering to some degree. Each rule may trigger ... |

220 |
Classifier systems and genetic algorithms
- Booker, Goldberg, et al.
- 1990
(Show Context)
Citation Context ...n of ELF to the development of Fuzzy Logic Controllers for autonomous agents and for the classical cart-pole problem. INTRODUCTION Genetic Algorithms (GAs)[13] and Learning Classifier Systems (LCS)[7]=-=[8]-=- emerged in the last years as powerful Evolutionary Learning (EL) techniques to identify systems that optimize some cost function. The cost function provides a reinforcement that guides the selection ... |

117 |
Knowledge Growth in an Artificial Animal
- Wilson
- 1985
(Show Context)
Citation Context ...ate is decreased. The goal is to obtain the minimum number of rules with a satisfactory performance. When the system is in a state not sufficiently covered with by any rule, a cover detector operator =-=[25]-=-[12] generates a new rule, having the antecedent that best matches the current state, and possibly containing some "don't cares". Therefore ELF may either build a rule base from scratch, or work with ... |

100 |
Fuzzy logic synthesis with genetic algorithms
- Thrift
- 1991
(Show Context)
Citation Context ...ITHMS AND FUZZY SYSTEMS Since 1989 [15], GAs have been adopted to identify sub-optimal FLCs [20][18][16][19]. There have been also proposals to extend the LCS approach to Fuzzy Classifier Systems [21]=-=[22]-=-[23]. Learning could be applied to at least 3 different aspects of an FLC: concept definition, state relevance, and the relationship between state and action. Concept definition People tend to describ... |

94 | Integrating design stages of fuzzy systems using genetic algorithms
- Lee, Takagi
- 1993
(Show Context)
Citation Context .... We also show some applications of ELF to conceptually interesting problems. EVOLUTIONARY LEARNING AGORITHMS AND FUZZY SYSTEMS Since 1989 [15], GAs have been adopted to identify sub-optimal FLCs [20]=-=[18]-=-[16][19]. There have been also proposals to extend the LCS approach to Fuzzy Classifier Systems [21][22][23]. Learning could be applied to at least 3 different aspects of an FLC: concept definition, s... |

88 |
Design of an adaptive fuzzy logic controller using a genetic algorithm
- Karr
- 1991
(Show Context)
Citation Context ...Algorithms has been addressed by many people (e.g., [20][18][16]. In some of these approaches the set of rules is defined a priori and the GA is used to optimize the shape of the membership functions =-=[14]-=-. In this case the GA tunes the FLC. The same can be said for those systems where fuzzy rules have a specific set of membership functions each [18][21]. In other approaches, (e.g., [20]) rules are gen... |

81 |
Fuzzy Control of pH Using Genetic Algorithms
- Karr, Gentry
- 1993
(Show Context)
Citation Context ... also show some applications of ELF to conceptually interesting problems. EVOLUTIONARY LEARNING AGORITHMS AND FUZZY SYSTEMS Since 1989 [15], GAs have been adopted to identify sub-optimal FLCs [20][18]=-=[16]-=-[19]. There have been also proposals to extend the LCS approach to Fuzzy Classifier Systems [21][22][23]. Learning could be applied to at least 3 different aspects of an FLC: concept definition, state... |

62 |
The Fuzzy Classifier System: a classifier System for Continuously Varying Variables
- Valenzuela-Rendon
- 1991
(Show Context)
Citation Context ...S AND FUZZY SYSTEMS Since 1989 [15], GAs have been adopted to identify sub-optimal FLCs [20][18][16][19]. There have been also proposals to extend the LCS approach to Fuzzy Classifier Systems [21][22]=-=[23]-=-. Learning could be applied to at least 3 different aspects of an FLC: concept definition, state relevance, and the relationship between state and action. Concept definition People tend to describe be... |

40 |
A new approach to fuzzy classifier systems
- Parodi, Bonelli
- 1993
(Show Context)
Citation Context ...AGORITHMS AND FUZZY SYSTEMS Since 1989 [15], GAs have been adopted to identify sub-optimal FLCs [20][18][16][19]. There have been also proposals to extend the LCS approach to Fuzzy Classifier Systems =-=[21]-=-[22][23]. Learning could be applied to at least 3 different aspects of an FLC: concept definition, state relevance, and the relationship between state and action. Concept definition People tend to des... |

31 |
Improved Fuzzy Process Control of Spacecraft Autonomous Rendezvous Using a Genetic Algorithm
- Karr, Freeman, et al.
- 1989
(Show Context)
Citation Context ...), an EL system that faces efficiently some of these problems. We also show some applications of ELF to conceptually interesting problems. EVOLUTIONARY LEARNING AGORITHMS AND FUZZY SYSTEMS Since 1989 =-=[15]-=-, GAs have been adopted to identify sub-optimal FLCs [20][18][16][19]. There have been also proposals to extend the LCS approach to Fuzzy Classifier Systems [21][22][23]. Learning could be applied to ... |

28 |
Behavior analysis and training: A methodology for behavior engineering
- Colombetti, Dorigo, et al.
- 1996
(Show Context)
Citation Context ..., without any need for learning. Unfortunately, this is not true for many real-world environments. Moreover, the verification of the correctness of a reinforcement program is still an open problem [3]=-=[9]-=-. We have found support to this opinion about the difficulty to evaluate the correctness of a reinforcement program in different experiments with autonomous agents [2][3][5], and other fuzzy control a... |

27 | The Evolution of Strategies for Multiagent Environments
- Grefenstette
- 1992
(Show Context)
Citation Context ...is decreased. The goal is to obtain the minimum number of rules with a satisfactory performance. When the system is in a state not sufficiently covered with by any rule, a cover detector operator [25]=-=[12]-=- generates a new rule, having the antecedent that best matches the current state, and possibly containing some "don't cares". Therefore ELF may either build a rule base from scratch, or work with an i... |

22 |
ELF: Learning incomplete fuzzy rule sets for an autonomous robot
- Bonarini
- 1993
(Show Context)
Citation Context ...ally compete with each other, since they propose different actions for the same state. Therefore, a possible solution to this problem is to have a population of rules partitioned into sub-populations =-=[2]-=-. Each sub-population contains rules that have the same antecedents, thus they match the same fuzzy state. Rules in a subpopulation compete with each other. Rules belonging to different sub-population... |

15 | Learning to to coordinate fuzzy behaviors for autonomous agents
- Bonarini, Basso
- 1997
(Show Context)
Citation Context ...s still an open problem [3][9]. We have found support to this opinion about the difficulty to evaluate the correctness of a reinforcement program in different experiments with autonomous agents [2][3]=-=[5]-=-, and other fuzzy control applications, such as: the cart-pole balancing [14][17], the spacecraft autonomous rendez-vous [15] and the truck backing-up [17]. We have found that it is common to have cri... |

14 |
Fuzzy Q-learning and dynamical fuzzy Q-learning
- Glorennec
- 1994
(Show Context)
Citation Context ...h a discount factor defined by the designer) rules triggered during past episodes. This contributes to evolve chains of rules, analogously to what happens with Q-learning [24] and fuzzy Q-Learning [1]=-=[11]-=-. The aspects of ELF common to these approaches are discussed in [6]. ELF reinforces rules that belong to a sub-population considering some estimate of the performance they may be expected to have. Th... |

13 |
A self-tuning method of fuzzy reasoning by genetic algorithm
- Nomura, Hayashi, et al.
- 1992
(Show Context)
Citation Context ...lems. We also show some applications of ELF to conceptually interesting problems. EVOLUTIONARY LEARNING AGORITHMS AND FUZZY SYSTEMS Since 1989 [15], GAs have been adopted to identify sub-optimal FLCs =-=[20]-=-[18][16][19]. There have been also proposals to extend the LCS approach to Fuzzy Classifier Systems [21][22][23]. Learning could be applied to at least 3 different aspects of an FLC: concept definitio... |

12 |
Fuzzy Q-Learning: a new approach for fuzzy dynamic programming
- Berenji
- 1994
(Show Context)
Citation Context ...with a discount factor defined by the designer) rules triggered during past episodes. This contributes to evolve chains of rules, analogously to what happens with Q-learning [24] and fuzzy Q-Learning =-=[1]-=-[11]. The aspects of ELF common to these approaches are discussed in [6]. ELF reinforces rules that belong to a sub-population considering some estimate of the performance they may be expected to have... |

12 |
Fuzzy Network Synthesis with Genetic Algorithms
- Feldman
- 1993
(Show Context)
Citation Context ...rresponds to learning the relevant configurations of the input values. Usually, the approaches concerning Evolutionary Learning of FLCs consider that the FLC covers all the input space (e.g., [22][18]=-=[10]-=-). In an FLC, the number of 2srules covering all the possible combinations of the antecedent values is equal to the product the number of possible values of each antecedent. For three variables with t... |

11 | The fuzzy classifier system: A classifier system for continuously varying variables - Valenzuela-Rend��n - 1991 |

8 | Genetic algorithms for the development of fuzzy controllers for autonomous guided vehicles - Leitch, Probert - 1994 |

6 | Some methodological issues about designing autonomous agents which learn their behaviors: the ELF experience
- Bonarini
- 1994
(Show Context)
Citation Context ...ior, without any need for learning. Unfortunately, this is not true for many real-world environments. Moreover, the verification of the correctness of a reinforcement program is still an open problem =-=[3]-=-[9]. We have found support to this opinion about the difficulty to evaluate the correctness of a reinforcement program in different experiments with autonomous agents [2][3][5], and other fuzzy contro... |

5 |
ZCS: A zeroeth level classifier system
- Wilson
- 1994
(Show Context)
Citation Context ... interferes with the evolution of specialized rules. The solutions we have adopted to solve these problems can also be extended to similar problems arising with other EL Algorithms, in particular LCS =-=[26]-=-[27]. Cooperation and competition Usually, the action an FLC produces for the actuators does not depend on one rule only, but on different rules, each one triggering to some degree. Each rule may trig... |

4 |
Classifier systems that learn their internal models
- Booker
- 1988
(Show Context)
Citation Context ...tion of ELF to the development of Fuzzy Logic Controllers for autonomous agents and for the classical cart-pole problem. INTRODUCTION Genetic Algorithms (GAs)[13] and Learning Classifier Systems (LCS)=-=[7]-=-[8] emerged in the last years as powerful Evolutionary Learning (EL) techniques to identify systems that optimize some cost function. The cost function provides a reinforcement that guides the selecti... |

2 |
Adaptive fuzzy model-following control: a robotic manipulator application
- Bonarini, Danieli, et al.
- 1994
(Show Context)
Citation Context ...values coming from his or her experience. We are also investigating the possibility to learn with ELF the structure of a FLC, and then optimize the membership function with another learning algorithm =-=[4]-=-, eventually cycling in this double learning activity. We are currently working in different directions. We are investigating the possibilities of ELF in dynamic environments, where antecedents can ma... |

2 |
Extending Q-learning to fuzzy classifier systems
- Bonarini
- 1995
(Show Context)
Citation Context ...ast episodes. This contributes to evolve chains of rules, analogously to what happens with Q-learning [24] and fuzzy Q-Learning [1][11]. The aspects of ELF common to these approaches are discussed in =-=[6]-=-. ELF reinforces rules that belong to a sub-population considering some estimate of the performance they may be expected to have. This solves the imperfect reinforcement program problem, discussed abo... |