FIG. 3.
The effect of single-neuron position and clutter sensitivity on population recognition performance. A: population performance on the recognition tasks with visual scenes containing single objects. Performance was averaged over multiple simulation runs; error bars indicate SD. - - -, the performance from shuffled runs (i.e., chance). The performance of the invariant populations performed above chance on the position-specific task because the neurons were sensitive to object identity and therefore conveyed some information about this conjoint identity and position task. B: example populations illustrating different amount of single-neuron position sensitivity. Each column is an example population consisted of neurons with a particular σp. Within each column, the top plot shows the responses of all the units to their most preferred object across changes in that object's position. The bottom panel shows the responses of an example unit to 3 different objects. The shape selectivity of all neurons was the same (i.e., same σs). C: population performance on visual scenes containing multiple objects. Different colors represent data from populations with different single-neuron clutter sensitivity [blue, complete clutter invariant (CCI); red, linear (LIN); green, average (AVG); magenta, divisive normalization (DIV)]. Because the simulation parameters and populations were not exactly matched, one should not make direct comparison of the absolute performance between A and C. The performance obtained using σp = 0.3 is shown in inset for better comparison. D: an illustration of a single-unit's responses to single objects and pairs of objects under different clutter rules.