Email updates

Keep up to date with the latest news and content from BMC Neuroscience and BioMed Central.

This article is part of the supplement: Eighteenth Annual Computational Neuroscience Meeting: CNS*2009

Open Access Poster presentation

Affect-driven learning in an avalanche neural network modeling early sensorimotor intelligence

Richard E Hill*, Quinn Macpherson and Richard B Wells

Author Affiliations

MRC Institute, University of Idaho, Moscow, ID, 83844-1024, USA

For all author emails, please log on.

BMC Neuroscience 2009, 10(Suppl 1):P143  doi:10.1186/1471-2202-10-S1-P143

The electronic version of this article is the complete one and can be found online at: http://www.biomedcentral.com/1471-2202/10/S1/P143


Published:13 July 2009

© 2009 Hill et al; licensee BioMed Central Ltd.

Poster presentation

This paper discusses a key learning element in an agent-based model of early sensorimotor learning during the first month of life. The system is based on the hypothesis that the earliest learned habits are driven by affectivity (innate non-objective preferences) and are grafted onto innate sensorimotor habits. The agent demonstrates the ability for learned associations, gradually overriding innate reflexes and producing learned voluntary motions. Additionally, the agent exhibits learned avoidance behaviors that could override innate affective preferences and reflexes [1].

The key learning element discussed here is an enhanced three-layer avalanche chain network. Avalanche chain neural network theory [2] was originally developed by Grossberg to model motor learning based upon a psychological model. He presented a conceptual mathematical schema for these systems. In this work, we report a specific implementation of the system. One finding of this work is that the original 2-level Grossberg topology must be extended to a 3-level system to deal with interruptions and unanticipated changes in the affective state of the agent. Specific sets of mathematical equations implementing robust avalanche chain learning are presented.

The avalanche network proxies functions of the motor cortex and lower level motor networks (e.g. in brain stem). Its learning dynamics are aroused by signals generated in a proxy limbic system and associated dipole-network conditioning functions [2]. Pre-set innate motor reflexes supply the unconditioned stimulus and learned responses are formed by mimicking these primitive reflexes (see Figure 1). Over time, the learned associations gradually take over control of motor movement execution by means of a simple assimilation network. This paper deals primarily with the learning dynamics of the 3-level avalanche.

thumbnailFigure 1. Block diagram of learning agent.

Acknowledgements

This work was supported by the National Science Foundation, award # 0648202.

References

  1. Hill R, Macpherson Q, Wells R: Conditioned learning dynamics in avalanche chain neural networks.

    1st INCF Congress on Neuroinformatics 2008 Abstracts, "Databasing and modeling the brain," International Neuroinformatics Coordinating Facility, Stockholm, Sweden, Sept. 7–9 2008, 108. OpenURL

  2. Grossberg S: A theory of human memory: Self-organization and performance of sensory-motor codes, maps, and plans. In Progress in theoretical biology. Volume 5. Edited by Rosen R, Snell S. New York: Academic Press; 1978::233-374. OpenURL