Abstract
Recent evidence suggests that visual information in memory can be stored as a hierarchy of representations from feature to object to gist of many objects (Brady & Alvarez, 2011; Brady et al., 2011, 2018). Memory for individual objects’ color is biased toward the gist in long-term memory, that is, mean color of a list of learned objects (Brady et al., 2018). Here we asked whether individual objects with outstanding features can affect the gist memory. Evidence from ensemble summary statistics suggests that salient or outstanding objects would either disproportionately influence the gist (Kanaya et al., 2018) or would be completely excluded from the gist (Haberman & Whitney, 2010).
In our experiment, we compared color gist-based bias in long-term memory for categories with and without outliers using the continuous report paradigm (Wilken & Ma, 2004; Zhang & Luck, 2008). The experiment consisted of five identical blocks. In each block, participants were asked to memorize 40 images of real-world objects from four different categories, such as armchairs, scarves, or notebooks. Individual object colors for each category were generated from a normal distribution with a randomly chosen mean from a CIE Lab color wheel. In half of the categories across all blocks, one object was a color outlier: Its color was located fairly at “tails” of the categorical normal distribution. The participants were then tested with 2AFC for recognition memory with old and new exemplars from the same categories. The participants had to choose an old exemplar and report its color using a color wheel. Color reports were analyzed using the mixture models. Our results showed that individual memories were biased towards the categorical mean, with outliers excluded from the mean (gist) computation. This supports the Bayesian integration model as an adaptive mechanism in visual long-term memory (Brady, Schacter, & Alvarez, 2018).