Abstract
Visual spatial attention has been shown to modulate the receptive fields (RFs) of cells in cortical area MT (Womelsdorf et al., Nat Neurosci, 2006, Anton-Erxleben et al., Cereb Cortex, 2009). Former models of attention are typically too qualitative or too simple (e.g. Gaussian model by Womelsdorf et al., J Neurosci, 2008) to capture all observations made. Thus, we mathematically derived a neuro-computational model of interacting MT neurons and fitted the free parameters to the individual electrophysiological cell measurements from macaque area MT of Womelsdorf et al. (2006) and Anton-Erxleben et al. (2009). The model has been derived from a biologically plausible representation of neuronal connections, whilst considering the excitation from the stimulus as well as the lateral divisive-inhibition from MT-cells responding to two distractors, additionally presented in the RF, moving in the anti-preferred direction of the recorded cell. The RF is parameterized as an elongated Gaussian shape. Attention is propagated multiplicatively to all MT-cells in a certain range according to a radial Gaussian function. The proposed divisive-inhibition model is showing effects of attentional shift, shrinkage and expansion when attending inside or close to the receptive field based on the strength and width of attention. Due to these effects we simulated a virtual "compression" experiment (Ross et al., Nature, 1997, Kaiser & Lappe, Neuron, 2004) using all fitted model neurons of the MT cells and setting the attention signal to a single location for all cells (the virtual saccade target). We then decoded the population response of all model MT cells for position in the attended and non-attended case, given a flashed stimulus. If we compare the attended and non-attended case we observe the typical "compression" pattern of mislocalization, which suggests that "compression" emerges by the attentive tuning of the neural response of a large number of individual cells.
Meeting abstract presented at VSS 2016