Modeling attentional modulated spike count correlation in macaque V1.

Kai Chen, Songting Li, Douglas Zhou

Please check the link for PDF version of poster (link to CCCN2021).

Keywords: Visual Attention, V1, Circuitry Rate Model, Differential Evolution, Mean Field Analysis

Abstract: Attention, including spatial attention and context-dependent attention, has been known as an essential cognitive function facilitate information processing through the resource-limited sensory pathways for human and non-human primates. The underlying neural mechanisms remain unknown yet and have been widely studied. According to previous experimental works for visual attention, it, on the one hand, facilitates the activation of target neurons, and on the other hands, decreases the spike-count correlation within target neuronal populations. Recently, our experimental cooperators investigated the macaque V1 neurons and found that attention can either increase or decrease neuronal responses. More interestingly, spike-count correlations between non-suppressive neuronal pairs(neurons are either attentional enhanced or attentional suppressed) decrease w.r.t. increment of attentional load(or task difficulty), while those of suppressive pairs (at least one neuron is attentional suppressive) increase w.r.t. growing attentional load. The conventional hypothesis for context-dependent attention believes that attentional signals act as top-down inputs from high-order cortical area to the primary sensory cortex. To verify the hypothesis, as well as understand the these nontrivial attentional modulation towards spike count correlations, we built a circuit rate model of neuronal populations based on experimental constraints to fit the activity of three types of attentional modulated neurons as the first step. Then by applying gradient-free optimization algorithm, differential evolution, we tuned the model parameters to fit the experimental constraints about firing rate and spike count correlation levels. Our model showed that inhibitory projections play vital role. Meanwhile, our model can flexibly predict the modulation of spike count correlation under different types of input drives as well as input contrasts. With mean field analysis, we further verify the range of parameter that acquired for the monotones of spike count correlation modulation. More importantly, our rate model can provide the potential tuition towards the potential candidate for those different types of attention modulated neuron types. Based on such intuition, we can further build spike neuronal network models to verify our hypothesis towards the top-down attentional mechanism. As an ongoing project, our work provide a framework to efficiently modeling attentional modulation towards spike count correlation with rate-based circuit model. With the gradient-free optimization for parameter tuning and mean field analysis, our framework provides sufficient intuitions about the potential mechanism of modulation.