Purchase this article with an account.
Paul Bays, Robert Taylor; Theory of neural coding predicts an upper bound on estimates of memory variability. Journal of Vision 2019;19(10):203b. doi: https://doi.org/10.1167/19.10.203b.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
A popular method for interpreting responses in memory tasks is to statistically decompose errors into a mixture of remembered (normally distributed) and guessing (uniformly distributed) components. While intuitive, models based on such discrete memory states have typically provided a worse fit to data than those based on a continuum of representational fidelity. In particular, a model based on storing information in a noisy population of idealized neurons provides a more parsimonious account of working memory errors while also having some correspondence with neurophysiological principles. Here we consider how best to interpret results obtained from the mixture method if the population coding account is correct. We show that for an idealized homogeneous neural population, the width of the fitted normal distribution cannot exceed the average tuning width of the component neurons, and that this holds to a good approximation for more realistic populations also. Examining eight published studies of orientation recall, we find a consistent pattern of results suggestive of an upper limit of approximately 20 degrees, which compares well with electrophysiological estimates of orientation tuning in visual cortex. We discuss the implications for previous studies that have interpreted a plateau in width of the normal component as evidence for limits on the precision of perception, working memory and long-term memory.
This PDF is available to Subscribers Only