Back
to indexFilippo Menczer, Richard K. Belew and Wolfram Willuhn
Summary
This paper presents the implementation of an adaptive agent, and provides experimental evidence to highlight the agent's efficiency. The intuition behind the agent is to pick up the query where a search engine left off. The results highlight the population dynamics involved in such a task, using endogenous fitness models as well as local and lookahead cloning. These algorithms provide a threshold from which offspring may be cloned.
This work begins a foundation for the InfoSpiders project, which allows adaptive agents to search the web based on a query. The utility of the endogenous fitness models is proven, and may continue to be the basis of further work.
Keywords
endogenous fitness, local cloning, lookahead cloning, adaption, distributed, lookahead, user feedback, on-line
Methods
The general algorithm consists of local selection of which agents to clone to increase the population, then local or lookahead cloning is used if the energy of the agent is sufficiently above a given threashold. Which link to follow next is determined through a probability distribution based on the relevance of the current document.
Rating
7
Bibtex Entry
@inproceedings = { menczer95,
author = "Filippo Menczer and Richard K. Belew and Wolfram Willuhn",
title = "Artificial Life Applied to Adaptive Information Agents",
booktitle = "{AAAI} Spring Symposium on Information Gathering",
year = "1995",
url = "citeseer.nj.nec.com/menczer95artificial.html"
}