Analysis of Dynamics and Object Recognition Performance in Coupled Map Networks
Coupled logistic map lattices can perform object recognition by mapping class members to a point in a space defined by partition cell occupancies, measured at a readout time t for the entire lattice. Each map (unit) represents average spiking dynamics in mixed excitatory-inhbitory neuronal populations. Non-stationary parameters improve peak recognition rate and recognition time, suggesting one role for interaction of multiple time scales in neural oscillations. A hypothesis that recognition functions by modulating the approach to an invariant measure was rejected by examining the response to noise. Moderate correlation (r=.3) between a configuration entropy measure and recognition rate is found