AIM: A network model of attention in auditory cortex
Fig 6
Functional implications of the AIM network: Spatial tuning of the network is dictated by the state of TD neurons.
(A) The network monitors the entire azimuthal plane when all TD neurons are active. (B) The network attends to a specific direction if the corresponding TD neuron is off. (C) The network attends to a different location if a different TD neuron becomes inactive. Column 3 shows the result of simulations when speakers are presented sequentially to the network, in spike rasters. Column 4 shows the result of simulations when speakers are presented simultaneously to the network. (D) The AIM network can recover an attended target within a speech mixture, as quantified by the cross-correlation measures between the simultaneous simulation network output and single speaker spike rasters. Error bars show standard deviation (n = 20). X-axis is the reference speaker, and each line color denotes the attended location. (E) Spatial separation of two talkers (S1, S2) vs. network performance, as quantified by correlation between the network output to the attended target. The encoding becomes more representative of the attended speaker as the separation between the two speakers increases. In the “not attending” case, S1 is used as the reference for correlation calculation. Shaded area represent 95% confidence interval, n = 20.