Purchase this article with an account.
Jochen Braun, Gustavo Deco, Alexander Pastukhov; Constraining the dynamics of multi-stable perception. Journal of Vision 2013;13(9):535. doi: https://doi.org/10.1167/13.9.535.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The dynamics of multi-stable perception is known to reflect several stabilizing and destabilizing factors. This interplay is captured by simplistic computational models (e.g., Laing & Chow, 2002, J. Comp. Neurosci., 'LC-model') with mutual inhibition, adaptation, and noise, at least under restrictive stimulus conditions. Hoping to constrain the parameters of such models, we collected reversal statistics from 24 observers and with 3 multi-stable displays (binocular rivalry, structure-from-motion, and Necker cube). We compared these observations to reversal sequences generated by a reduced LC-model in a 5-dimensional parameter space. As expected, the distribution of dominance periods exhibited a highly consistent shape (Gamma-like with α=3.6±0.5), which has been proposed as an effective constraint for multi-stable dynamics (Shpiro et al., 2009, J. Comp. Neurosci.). However, we found that approximately 66% of the possible parameter volume generated realistic distribution shapes with suitable levels of noise. Thus, distribution shape is not, by itself, an effective constraint. Also as expected, reversal sequences exhibited weak (up to c=0.4) but significant correlations between dominance episodes and an integral measure of prior perceptual history ('cumulative history', Pastukhov & Braun, 2011, J. Vis.). In addition, serial correlations (up to c=0.2) between successive dominance episodes also reached significance occasionally (Van Ee, 2009, J. Opt. Soc. Am. A). We found that both observations are consistent with and, indeed, predicted by, the LC-model with suitable levels of adaptation. When combined, dominance distribution and history-dependence constrained multi-stable dynamics effectively. Approximately 15% of the possible parameter volume of the LC-model generated reversal sequences with realistic statistics in both regards. Between 3% and 6% of the parameter volume reproduced the perceptual dynamics of individual observer/display combinations. We conclude that the combination of dominance distribution and history dependence constrains the perceptual dynamics of individual human observers, revealing the operative balance of mutual inhibition, adaptation, and noise.
Meeting abstract presented at VSS 2013
This PDF is available to Subscribers Only