Angus Chapman¹, Rachel Denison¹; ¹Boston University

Motivation: Visual attention can be deployed to spatial locations or points in time, resulting in enhanced processing of attended information and impaired processing of unattended information. Such tradeoffs are thought to reflect competition among stimuli for limited attentional resources. Normalization-based models implement attention by enhancing neural excitation prior to divisive suppression, but do not explain why tradeoffs arise. Here, we tested whether normalization, when coupled with an optimal attentional allocation policy, could incentivize tradeoffs in attentional allocation without appeal to an explicit limited attentional resource. Methods: We adapted a dynamic spatiotemporal normalization model to incorporate attentional modulation that varied across space and time. Spatiotemporal normalization was implemented through neurons with spatiotemporal receptive fields, with excitatory and suppressive drives that depended on inputs pooled across space and time. We simulated orientation discrimination performance for two stimuli appearing at varying locations and times, with different degrees of proximity, and 75% valid spatial or temporal cues. Attentional gain was applied independently to both stimuli, without limit, but adjusted to optimal levels that maximized the model’s overall performance. Results: Across a range of conditions, model performance was best when attentional gain was allocated selectively to the cued stimulus, even compared to maximally attending both stimuli. Thus, the optimal attention policy led to focused attention in space and time. Importantly, these tradeoffs only occurred when stimuli were close enough in space or time that they mutually suppressed each other via normalization. Additionally, when stimuli were closer in spatial proximity, it was optimal to direct attention slightly away from the cued location, minimizing simultaneous enhancement of the uncued stimulus. Conclusions: In the presence of normalization, attentional tradeoffs emerge from the optimal allocation of attention in space and time, with no attentional resource limit. Thus, normalization suffices to incentivize focused spatial and temporal attention.

Acknowledgements: This work was supported by National Institutes of Health National Eye Institute (R01 EY037358) and BU startup funding to R.N.D.