The Poisson shot noise model of visual short-term memory and choice response time: Normalized coding by neural population size

A normalized coding condition is proposed that provides a theoretical link between the Poisson shot noise model of choice response time and a Poisson neuron model of the information capacity of visual short-term memory (VSTM). In both models, noise in the cognitive representations of stimuli is attributed to Poisson variability in the neural processes that encode them. In VSTM, a Poisson coding model predicts the invariance of ∑i(di′)2 across displays of different sizes, as is found experimentally, but it incorrectly predicts that the diffusion coefficient of the approximating diffusion process will decrease with its drift rate. Normalized coding assumes that the squared magnitudes of the random perturbations to a stimulus representation are inversely proportional to the number of Poisson neurons that represent it. The normalized coding model correctly predicts both a constant diffusion coefficient and the invariance of ∑i(di′)2, as required by the experimental data. Normalized coding reconciles the theoretical and empirical properties of diffusion models of decision-making and the sample-size model of VSTM.