Abstract
Binding refers to the information that groups different features together into objects. In probed recall tasks, binding failure is associated with swap errors, in which observers report an item other than the one indicated by the probe. To investigate the basis of swap errors, we presented observers with a memory display consisting of six coloured discs. After a delay they were asked to report, on a continuous scale, either the colour they had seen at a specified location, or the location that corresponded to a specified colour. Swap errors were observed to occur most frequently for items that were similar to the target in the probe feature-dimension, while swap frequency fell to chance for the most dissimilar items, indicating that variability in memory for the probe feature was responsible for swaps. These observations were reproduced by a model of population coding, in which visual information was encoded in the noisy activity of a single population of neurons with conjunction (colour × location) responses. Maximum likelihood decoding of population activity resulted in errors that quantitatively reproduced the behavioural distributions of recall error, the frequency of swaps, and the relationship between swap frequency and probe-dimension similarity. These results demonstrate a neural architecture that could underlie binding in working memory.
Meeting abstract presented at VSS 2016