• Summary
• Related Documents
  • Active Bibliography
  • Co-citation
• Version History

Abstract:

If behavior networks, which use spreading activation to select actions, are analogous to connectionist methods of pattern recognition, then recurrent behavior networks, which use energy minimization, are analogous to Hopfield networks. Hopfield networks memorize patterns by making them attractors. Similarly, each behavior of a recurrent behavior network should be an attractor of the network, to inhibit fruitless, repeated switching between different behaviors in response to small changes in the environment and in motivations. I overcome two major objections to this view, and demonstrate that the performance in a test domain of the Do the Right Thing recurrent behavior network is improved by redesigning it to create desirable attractors and basins of attraction. I further show that this performance increase is correlated with an increase in persistence and a decrease in undesirable behavior-switching. On a more general level, this work encourages the study of action selection as a dynamics problem. Key Words: action selection; decision making; attractors; behavior networks; pattern recognition; nonlinearity ii Acknowledgments My advisor, Deborah Walters, gave me guidance and encouragement, and was on my side throughout the writing of this dissertation. William Rapaport advised me in the early stages of this work, and made many helpful suggestions that improved this document's readability and forcefulness. Patricia Eberlein helped me through several unpleasant mathematical quandaries. Erwin Segal was my outside reader on short notice. I thank the SNePS Research Group for ideas about knowledge representation, reasoning, and acting, that, although they did not appear in this research, played a major part in defining its direction. I also want to thank Pattie Maes and Bradley Rhodes of MIT for e-mail discussions of behavior networks, and especially Toby Tyrrell of the Southampton Oceanography Centre for making his software available and for many lengthy e-mail discussions concerning his work. Comments by

Citations