Abstract
Attention modulates visual responses in diverse ways, affecting not only firing rates, but also the position, size and structure of receptive fields (shifting or extending receptive fields towards the focus of attention, both spatially and feature-wise). While recently introduced models can explain many of the effects of attention on firing rates, no current model can simultaneously account for the shifting, scaling and retuning of receptive fields reported by recent studies. Here we suggest that all these effects arise naturally from a top-down modulation, targeting the top-most areas in the hierarchy (e.g. V4 or MT), being propagated down towards lower areas through feedback connections, and interacting with short-range mutual inhibition between neighboring neurons. While feedback connections redistribute the modulation to inputs (explaining long-distance effects on receptive field position and tuning), short-range inhibition produces competition between stimuli, the range of which is automatically scaled to the size of receptive fields, in accordance with experimental results. We use a simple two-layer model, with reciprocal connections between layers and mutual inhibition between neighboring neurons, in which bottom-layer neurons are tuned to oriented edges, and top-layer neurons pool inputs from similarly-tuned neurons over a limited region of the bottom layer; attention is introduced as an excitatory modulation targeting top-layer neurons only. We show that this simple model reproduces many of the known effects of attention on visual responses,including: (i) biased competition, (ii) contrast or response gain, (iii) sharpening or scaling of tuning curves, as well as (iv) the shifting, scaling and spectral retuning of receptive fields. Our results suggest a “null model” for attentional effects in the visual cortex, in which the existence of feedback connections and short-range mutual inhibition suffices to explain all known effects of attention on neural responses.