Behavior Self-Organization in Multi-Agent Learning.
Abstract
There are four primary results of the first year of the project: It was discovered that clustering algorithms for pre-sorting high-dimensional datasets was not effective in improving subsequent processing by reinforcement learning methods. It was discovered that Bayesian belief networks can be combined with decision nodes and an incremental assessment algorithm to mimic human patterns of data reduction and knowledge representation. The human immunological system was identified as a possible model for a "bidirectional" distributed decision network. Initial work has identified a model-balancing technique, borrowed from linear system theory, that is a strong candidate for a pruning and model reduction method for large modular networks.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 28, 1999
- Accession Number
- ADA366166
Entities
People
- Hugh F. Vanlandingham
- John S. Bay