Fig 2. Simple cue-probe experiment and model design with example raster plots.

A. Temporal structure of cue-probe trials. During each trial, the cue and probe were both presented for 500 ms at 500 and 3000 ms, respectively. B. Design of the ring attractor and important receptor subtypes for excitatory and inhibitory connections. The ring attractor consisted of 1024 pyramidal cells and 256 interneurons. The pyramidal cells were assigned radial directions and those near π/2 and 3π/2 were associated with the cue and probe stimuli, respectively. Connections indicated with a dashed line were spatially localized whereas those indicated with a solid line were not. C. Plot of the weight distributions of localized and global connections. Localized weights were more strongly connected to pyramidal neurons with a similar radial direction. The difference between global inhibitory and localized excitatory connections resulted in the creation of activity bump [23]. D. An example of AMPA, NMDA, and GABA receptor currents and how those manifested as spiking activity. In this example, EPSPs were induced at a rate of 20 hertz for the entire duration and IPSPs were induced at a rate of 20 hertz starting at 250 ms. The introduction of the IPSPs reduced the spiking rate (red diamonds) shown in the top panel as the excitatory and inhibitory currents interacted. Leak and noise currents are omitted. E-G. Example raster plots that show how activity bumps are induced and represent the cue and probe. E. With activity bump maintenance, the neural representation of the cue stayed at π/2 despite afferent signal at 3π/2 during the probe (NMDA_g,pyr = 0.37 μS, NMDA_g,int = 0.33 μS, AMPA_g,Aff = 0.9 μS). F. In the case of activity bump jumps, the neural representation switched from the cue to the probe (NMDA_g,pyr = 0.37 μS, NMDA_g,int = 0.30 μS, AMPA_g,Aff = 0.9 μS). G. The activity bump can also collapse prior to probe presentation (NMDA_g,pyr = 0.33 μS, NMDA_g,int = 0.30 μS, AMPA_g,Aff = 0.9 μS).