System identification in Priming of Pop-Out

Inter-trial repetitions of a target's features in a visual search task reduce the time needed to find the target. Here I examine these sequential dependencies in the Priming of Pop-Out task (PoP) by means of system identification techniques. The results are as follows. Response time facilitation due to repetition of the target's features increases linearly with difficulty in segmenting the target from the distracters. However, z-scoring the reaction times normalizes responses by equating facilitation across levels of difficulty. Memory kernels, representing the influence of the current trial on any future trial, can then be calculated from data normalized and averaged across conditions and observers. The average target-defining feature kernel and the target position kernel are well fit by a sum of two exponentials model, comprised of a high-gain, fast-decay component and a low-gain, slow-decay component. In contrast, the average response-defining feature kernel is well fit by a single exponential model with very low-gain and decay similar to the slow component of the target-defining feature kernel. Analysis of single participant's data reveals that a fast-decay component is often also present for the response-defining feature, but can be either facilitatory or inhibitory and thus tends to cancel out in pooled data. Overall, the results are similar to integration functions of reward history recently observed in primates during frequency-matching experiments. I speculate that sequential dependencies in PoP result from learning mechanisms that bias the attentional weighting of certain aspects of the stimulus in an effort to minimize a prediction error signal.