Purchase this article with an account.
Hsin-Hung Li, James Rankin, John Rinzel, Marisa Carrasco, David Heeger; An attention model of binocular rivalry. Journal of Vision 2017;17(10):579. doi: 10.1167/17.10.579.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Introduction: Binocular rivalry ceases when attention is diverted away from the stimuli (Zhang et al., 2011; Brascamp and Blake, 2012). Previous models do not explain this finding. We propose a computational model to address how the dynamics of rivalry depend on attention. Model: The sensory representation is composed of 4 monocular neurons (two eyes and two orthogonal orientations), 2 binocular-summation neurons that sum responses of the monocular neurons with the same orientation preference, and 4 binocular-opponency neurons that respond to conflicting information between eyes. Moreover, two 2nd-layer neurons, selective for orthogonal orientations, driven by the responses of the summation neurons provide feedback to the monocular neurons. Rivalry is driven by: (1) Slow adaptation in monocular and summation neurons. (2) Attention: Orientation-selective feedback from the 2nd-layer neurons, analogous to feature-based stimulus-driven attention, is active under the attended condition but silent under diverted attention. This feedback enhances the multiplicative gain of whichever orientation has stronger sensory responses, and suppresses the gain of the other. (3) Mutual inhibition: When active, opponency neurons provide subtractive suppression to the monocular population with weaker responses. Results: (1) The model exhibited rivalry for attended dichoptic gratings, but not for plaid or unattended stimuli. (2) When the two images were swapped between eyes rapidly, the dominant percept either stayed with one eye or followed one orientation for a period of time, depending on the stimuli's temporal characteristics. (3) The dominance duration changed with stimulus strength following Levelt's propositions. With a bifurcation analysis, we identified the parameter space in which the model's behavior was consistent with experimental results. Conclusion: In the model, attention affects rivalry by biasing competition toward the dominant feature, thereby stabilizing the percept. This is the first model that exhibits attention-dependent dynamics and captures a wide range of rivalry phenomena.
Meeting abstract presented at VSS 2017
This PDF is available to Subscribers Only