Abstract
The perception of body motion can show multi-stability with respect to the perceived walking direction(Vanrie, Dekeyser, Verfaillie, 2004). This phenomenon is not captured by existing neural models. Ipresent a new neurodynamical model that is based on a 2D neural field of snapshot neurons. encodingindividual views of keyframes of actions. These (nonlinear) neurons are coupled by lateral connections,which are symmetric with respect to neurons encoding different views, and asymmetric for neuronsencoding different snapshots, resulting in temporal sequence selectivity. In addition, the neurons adaptafter they have been activated, where the adaptation dynamics was fitted to electrophysiological data onrepetition-suppression in area IT (deBaene&Vogels, 2010). The model reproduces perceptual switchingand sequence selectivity of visual action recognition. In addition, quantitative fitting of neural andpsychophysical data leads to the following conclusions: 1) Perceptual switching in action perception isdriven by noise, not adaptation. 2) Consistent with recent data (Caggiano et al. 2013), the model predictscorrectly that repetition-suppression in action-selective neurons is much smaller than in shape-selectiveneurons. 3) It predicts a new action stimulus that should result in much stronger repetition-suppressionthan the stimuli used in previous physiological and imaging-experiments.
