Purchase this article with an account.
Shaul Hochstein, Noam Khayat, Marina Pavlovskaya, Stefano Fusi; How we perceive ensemble statistics and how they serve memory representation. Journal of Vision 2020;20(11):516. doi: https://doi.org/10.1167/jov.20.11.516.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Perceiving ensemble statistics, the mean and range of stimulus sets, has been the subject of great recent interest. Clearly, when there is too little time to see too much, perceiving the gist or Gestalt of a scene in terms of its statistics, facilitates overcoming visual system and memory capacity limits. It is now clear that we perceive stimulus set mean and range of size, brightness, color, orientation, facial emotion, lifelikeness, and known or novel category. Set perception is rapid, automatic, implicit and on-the-fly. But how can the brain compute the mean of a series of stimuli without representing the individual set members? We propose that the well-studied neural population code, coupled with ubiquitous broad receptive fields, solves this dilemma directly. The overlapping broad receptive fields for responses to set individuals prevents their individuation but expedites perception of their population mean. Only with focused attention can individuals be perceived. However, overcoming perceptual limitations is but half the story, because object representation and memory also have their limits, and even gradual accumulation of information can overload these systems. Benna and Fusi (2019) recently suggested that memory representation is compressed when related elements (called “descendants”) are represented in terms of their mean (progenitor or “ancestor”) and the difference of each from it. Applying this theory to ensemble statistics, we conclude that mean perception may be no more than an epiphenomenon of an essential generalized mechanism for memory compression. Implications will be demonstrated and discussed.
This PDF is available to Subscribers Only