Combining spatial and parametric working memory in a dynamic neural field model

Wojtak, Weronika; Coombes, Stephen; Bicho, Estela; Erlhagen, Wolfram

Abstract

We present a novel dynamic neural field model consisting of two coupled fields of Amari-type which supports the existence of localized activity patterns or “bumps” with a continuum of amplitudes. Bump solutions have been used in the past to model spatial working memory. We apply the model to explain input-specific persistent activity that increases monotonically with the time integral of the input (parametric working memory). In numerical simulations of a multi-item memory task, we show that the model robustly memorizes the strength and/or duration of inputs. Moreover, and important for adaptive behavior in dynamic environments, the memory strength can be changed at any time by new behaviorally relevant information. A direct comparison of model behaviors shows that the 2-field model does not suffer the problems of the classical Amari model when the inputs are presented sequentially as opposed to simultaneously.

Citation

Wojtak, W., Coombes, S., Bicho, E., & Erlhagen, W. (in press). Combining spatial and parametric working memory in a dynamic neural field model. Lecture Notes in Artificial Intelligence, 9886, https://doi.org/10.1007/978-3-319-44778-0_48

• Journal Article Type: Article
• Acceptance Date: Aug 13, 2016
• Online Publication Date: Aug 13, 2016
• Deposit Date: Feb 28, 2017
• Publicly Available Date: Feb 28, 2017
• Journal: Lecture Notes in Computer Science
• Electronic ISSN: 0302-9743
• Publisher: Springer Verlag
• Peer Reviewed: Peer Reviewed
• Volume: 9886
• Book Title: Artificial Neural Networks and Machine Learning – ICANN 2016
• DOI: https://doi.org/10.1007/978-3-319-44778-0_48
• Public URL: http://eprints.nottingham.ac.uk/id/eprint/40913
• Publisher URL: http://link.springer.com/chapter/10.1007/978-3-319-44778-0_48
• Copyright Statement: Copyright information regarding this work can be found at the following address: http://eprints.nottingham.ac.uk/end_user_agreement.pdf
• Additional Information: The final publication is available at Springer via http://dx.doi.org/10.1007/978-3-319-44778-0_48. Volume title: Artificial Neural Networks and Machine Learning – ICANN 2016