September 2021
Volume 21, Issue 9
Open Access
Vision Sciences Society Annual Meeting Abstract  |   September 2021
Behavioral benefits of spatial attention explained by multiplicative gain, not receptive field shifts, in a neural network model.
Author Affiliations & Notes
  • Kai Fox
    Stanford University
  • Dan Birman
    University of Washington
  • Justin Gardner
    Stanford University
  • Footnotes
    Acknowledgements  We acknowledge support from Research to Prevent Blindness and Lions Clubs International Foundation, and the Hellman Fellows Fund to JLG as well as the UW VTG (NEI T32EY07031) and WRF Postdoctoral Fellowship to DB. We thank Josh Wilson for help with data collection.
Journal of Vision September 2021, Vol.21, 1927. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Kai Fox, Dan Birman, Justin Gardner; Behavioral benefits of spatial attention explained by multiplicative gain, not receptive field shifts, in a neural network model.. Journal of Vision 2021;21(9):1927.

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

Spatial attention in humans and animals has been found to correlate with gain changes as well as with shifts in location and size of receptive fields in early visual cortex. But which, if any, of these physiological effects can account for the behavioral benefits of attention? Here we show in a computational model that changes in receptive field location and size are consequences of applying multiplicative gain and that they are not required for the behavioral benefits of attention. Instead, gain alone causes attended regions to have a larger impact on decision making. We asked human observers and a neural network model of the ventral visual stream to detect whether or not an object of a given category was present in a grid of four images. On focal attention trials, human observers were informed by a cue that the target could appear only at one location, improving performance. We found that applying Gaussian gain at the cued location in the early layers of the neural network model could mimic this cued performance improvement. Gain also caused receptive fields of later layers to decrease in size and shift towards the cued location. Through mechanistic models designed to isolate the behavioral consequences of each effect separately, we found that a gain applied early in the network increases the effective weight of the attended region at the last stage of the network, entirely explaining the behavioral benefits. In contrast, receptive field size and location changes in the absence of a gain component did not improve task performance. Our findings suggest that changes in receptive fields are epiphenomenal, unrelated to the causal changes that lead to behavioral improvement during directed spatial attention.


This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.