2019 Conference on Cognitive Computational Neuroscience

13-16 September 2019    Berlin, Germany

Technical Program

Paper Detail

Paper: PS-1B.69
Session: Poster Session 1B
Location: H Fläche 1.OG
Session Time: Saturday, September 14, 16:30 - 19:30
Presentation Time:Saturday, September 14, 16:30 - 19:30
Presentation: Poster
Publication: 2019 Conference on Cognitive Computational Neuroscience, 13-16 September 2019, Berlin, Germany
Paper Title: Spatial Attention introduces Behavioral Trade-off in a Large-Scale Spiking Neural Network
Manuscript:  Click here to view manuscript
License: Creative Commons License
This work is licensed under a Creative Commons Attribution 3.0 Unported License.
DOI: https://doi.org/10.32470/CCN.2019.1098-0
Authors: Lynn K. A. Sörensen, University of Amsterdam, Netherlands; Davide Zambrano, Centrum Wiskunde Informatica, Netherlands; Heleen A. Slagter, Vrije Universiteit Amsterdam, Netherlands; H. Steven Scholte, University of Amsterdam, Netherlands; Sander M. Bohté, Centrum Wiskunde Informatica, Netherlands
Abstract: Visuo-spatial attention is a key mechanism for selecting goal-relevant information in natural scenes. We here implement a variant of the normalization model of attention into a spiking convolutional neural network, which approximates attentional gain with a change in firing rates. We apply this type of attention with different spatial extents to various levels in the processing hierarchy of a network performing object recognition in natural scenes. We find that close to the average object-size attentional kernels yield the best performance, equivalent to a rather focused attentional enhancement. Furthermore, manipulating spatial attention within a single level was ineffective as benefits of spatial attention only arose from the combination of early-to-mid level modulations in the network hierarchy. Our results demonstrate that one can efficiently boost performance on the challenging task of recognizing objects in cluttered environments in a large-scale vision model by understanding attentional gain as a more or less precise representation of sensory information.