Technical Program

Paper Detail

Paper: PS-2B.7
Session: Poster Session 2B
Location: Symphony/Overture
Session Time: Friday, September 7, 19:30 - 21:30
Presentation Time:Friday, September 7, 19:30 - 21:30
Presentation: Poster
Publication: 2018 Conference on Cognitive Computational Neuroscience, 5-8 September 2018, Philadelphia, Pennsylvania
Paper Title: A flexible model of working memory
Manuscript:  Click here to view manuscript
Authors: Flora Bouchacourt, Tim Buschman, Princeton Neuroscience Institute, United States
Abstract: Working memory provides the workspace for holding and manipulating thoughts. It is flexible: we can hold anything in mind. However, typical models of persistent activity rely on tightly tuned attractors and do not allow for the flexibility observed in behavior. Here we present a novel model of working memory that maintains representations through random reciprocal connections between two layers of neurons: a selectively tuned layer and a randomly connected, untuned layer. As the recurrent interactions are unstructured, the network is flexible: it is able to maintain any input. However, adding multiple memories lead to interferences in the untuned layer, which result in a capacity limitation on the number of items that can be maintained. This is due to divisive-normalization-like reduction in neural responses coming from E/I balance in the network. Furthermore, it has been shown that time and load have a degrading effect on memory precision. Interferences in the network provide a possible mechanism for this psychophysical finding, as well as key neurophysiological results. Thus, we present a simple network model that allows for flexible representations while still capturing behavioral and neural hallmarks of working memory.