Two-dimensional Leaky Competing Accumulator Model for Saccadic Distribution Generation

The Leaky Competing Accumulator model (LCA) of Usher and McClelland is able to imitate the time course of perceptual decision making with use of arbitrary amount of visual stimuli as an input. We propose a new implementation called Spatial Leaky Competing Accumulator (sLCA), which exploits this LCA ability, but uses a saliency map as an input instead of multiple items abstract in space. The sLCA shares such parameters of the original model as leakage, recurrent self-excitation, randomness and non-linearity but also includes the novel version of lateral inhibition parameter. It allows only inhibition of immediate neighbors instead of all processing units, which is motivated by a better biological plausibility. All parameters of both original and improved models were optimized using the genetic algorithm, and their performance was compared in simulating human RT distribution in visual search task. According to Kolmogorov-Smirnov statistic, only the sLCA was able to fit the positively skewed human temporal data distribution.