Australasian Mathematical Psychology Conference 2019

Examination of doubly stochastic processes in a neural model of visual working memory.

Rob Taylor
Psychology, University of Newcastle

A popular method for probing the precision of visual working memory is to ask individuals to reproduce encoded stimulus features using a continuous response scale. A neural resource model based on storing information in a noisy population of idealized neurons provides a very parsimonious account of continuous report recall errors while also having some correspondence with neurophysiological principles. In this version of the model, random spiking in the neural population is typically approximated using a homogenous Poisson process. However, population activity is considered to be purely stimulus driven and thus ignores the possible contribution of non-sensory factors – such as arousal, attention, and adaptation - on cortical excitability. Here I provide a preliminary examination of a doubly stochastic neural resource model for the continuous report task. To account for non-sensory factors, the model generalises the Poisson process by introducing trial-by-trial fluctuations in neural gain. Initial assessment suggests that while the doubly stochastic model might provide an improved fit to behavioural data in some cases, model evaluation is complicated by the unrealistically low firing rates generated by the models.