A normative model of attention: receptive field modulation

When sensory stimuli are encoded in a lossy fashion for efficient transmission, there are necessarily tradeoffs between the represented fidelity of various aspects of the stimuli. In the model of attention presented here, a top-down signal informs the encoder of these tradeoffs. Given the stimulus ensemble and tradeoff requirements, our system learns an optimal encoder. This general model is instantiated in a simple network: an autoencoder with a bottleneck, innervated by a top-down attentional signal, and trained using backpropagation. The modulation of neural activity learned by this model qualitatively matches that measured in animals during visual attention tasks.