Abstract
How does the visual system choose among multiple percepts evoked by ambiguous or inconsistent stimuli? Spontaneous percept-switching under steady viewing is broadly explained by slow neural adaptation and noise disrupting fast cross-inhibition that stabilizes either percept in the short term, but this does not explain why interrupting the stimulus often causes the same percept to reappear across many ON/OFF-cycles. In fact, we find that most existing proposals for explaining such percept repetition actually predict the converse: stimulus-locked percept alternation.
We construct and analyze the simplest type of neural model that robustly captures both the spontaneous switching and the onset-induced choice process, and find that the choice dynamics depends crucially on how adaptation interacts with a hitherto neglected near-threshold neural input baseline (or slow coupling). Choice- and switching-events correspond to qualitatively different dynamical processes generated by a single, effectively 1-layer neural network (capable also of describing the percept-coding activity of several hierarchically coupled neural stages). Our model not only offers the first mechanistic explanation of how interruptions cause percept repetition, but also predicts other choice sequences. For example, stimulus-locked alternation of percepts is predicted for stimulus OFF-times well below the adaptation time. Our psychophysical experiment confirms this. No top-down signals are required to determine each choice of percept, but they may modulate parameters that shift the ON-OFF time regime in which the various sequences occur. The model also provides a mechanistic basis for ‘priming’ and ‘bottom-up attention’ effects on the resolution of perceptual ambiguities.