Caltech
Center for Neuromorphic Systems Engineering

Home
Research
News
People

[back]

Computational Modeling of Feature Inheritance
Whee Ky (Wei Ji) Ma

The proposition that reentrant interactions into the early visual system are necessary for visual awareness has lately been under close scrutiny. We examine this proposition in the context of a neuronal model which explains the phenomenology of feature inheritance.

Feature inheritance (Herzog and Koch) is a class of visual illusions in which a mask stimulus acquires a feature from a preceding brief target stimulus. We focus on the case where the target stimulus is a single tilted bar (on for 30 ms), and the mask stimulus is a grating of straight bars (on for 300 ms). In this case, only a tilted grating is perceived. If the mask is on for a short time (60 ms), target and mask are perceived superimposed.

Feature inheritance thus exhibits aspects of both backward masking and of temporal integration. Correspondingly, our model consists of two pathways: a multiplicative hypothesis-testing one and an activity-accumulating one. The multiplication pathway multiplies a "template" (a hypothesis about the world, based on early stimuli) with the current bottom-up input, in order to test the hypothesis. The template can be influenced by high-level expectations. If the match is good, the result is a bound object which is subsequently sent to perception. If it is bad, such as when the input is very brief or rapidly changing, the brain uses the output of the second pathway, which accumulates activity during the time the hypothesis-testing has not yet been completed. This yields temporal integration.

Although it is possible that the multiplicative pathway involves descending connections, no feedback in a graph-theoretical sense is necessary. Our model can explain much of the psychophysics of brief visual stimuli. In particular, we claim that the reconstruction of the contents of the perceptual gap due to the hypothesis-testing is exactly Efron's ``minimal perceptual moment''. Furthermore, our model is consistent with the computational model of object substitution masking as presented by DiLollo et al.

References
M.H. Herzog and C. Koch, Seeing properties of an invisible object: Feature inheritance and shine-through, Proc. Natl. Acad. Sci. USA 98 (2001), 4271

Di Lollo, J.T. Enns, and R.A. Rensink, Competition for Consciousness Among Visual Events: The Psychophysics of Reentrant Visual Processes, J. of Experimental Psychology 129 (2000), 481-507

 

top