Representational issues in machine learning
Classifier systems are numeric machine learning systems. They are machine counterparts to the natural genetic process and learn by reproduction, crossover, and mutation. Much publicity has been attended to their ability to demonstrate significant learning from a random start and without human intervention. Less well publicized is the considerable care that must be given to the choices of parameter settings and representation. Without the proper ''nurturing environment'' genetic algorithms are apt to learn very little. This infusion of human intelligence is often discounted, but the choice of appropriate representation forms the core of much of the current genetic algorithm research. This paper will address some of the representational issues from the perspective of two current experiments, one with scheduling and the other with a simulated robot. 10 refs., 7 figs.
- Research Organization:
- Oak Ridge National Lab., TN (USA)
- DOE Contract Number:
- AC05-84OR21400
- OSTI ID:
- 7190157
- Report Number(s):
- CONF-861096-3; ON: DE87004617
- Resource Relation:
- Conference: International symposium on methodologies for intelligent systems, Knoxville, TN, USA, 22 Oct 1986; Other Information: Paper copy only, copy does not permit microfiche production
- Country of Publication:
- United States
- Language:
- English
Similar Records
Emerging materials intelligence ecosystems propelled by machine learning
ISMB Conference Funding to Support Attendance of Early Researchers and Students