Quantitative characterization of energy landscapes in motion binding M Mattia1, G Aguilar2, A Pastukhov3, J Braun3 |
|---|
1Department of Technologies and Health, Istituto Superiore di Sanita, Italy |
| Visual perception exhibits numerous cooperative phenomena suggestive of attractor dynamics, such as order-disorder transitions or hysteresis (e.g. Buckthought et al, 2008, Vision Research, 48(6), 819-830). Here we ask whether the perception of coherent motion in random-dot kinematographs (RDK) is consistent with the dynamics of a cortical network model, specifically, with an input-dependent family of 'energy landscapes' governing the evolution of state trajectories. Six observers viewed RDK in which the fraction of coherent dots followed an unpredictable random walk and reported their initial and final percepts. The results revealed extensive path-dependence (hysteresis) of the final percept and a broad bistable regime for intermediate coherence fractions. The detailed information from random walk trials sufficed to constrain the first-order dynamical equation of a recurrently connected system (time-constant, non-linear feedback described by a general logistics function, and noise) and therefore revealed the energy landscape governing activity dynamics at each coherence level. Our analysis showed that hysteresis in the perception of coherent motion is consistent with bistability (and not with dynamical inertia) and, for the first time, quantitatively characterizes the 'basin of attraction' around a cooperative perceptual state. This opens novel perspectives for reverse-engineering the effective dynamical features of perceptual representations from non-stationary observations. |
| Up Home |
|---|