The Brain might Optimize Visual Acuity using Cortical Noise

Donald Glaser
Professor of the Graduate School Division of Neurobiology
Our goal is to construct computational models of the human visual system which explain its performance in terms of its physiology and anatomy.

In vision experiments we present images on computer monitors to subjects who are asked to judge shape, depth, color, velocity, texture, motion, etc. In a recent study of stereopsis, subjects could judge the relative depth of two adjacent test dots in the center of the screen very accurately. Next the "scene" was enclosed in a "picture frame" 50 degrees wide. When the frame was "tilted" in stereo, a systematic bias was discovered in the test dot judgments, even though the subjects didn't know that the frame was being manipulated and couldn't report it correctly. Thus a powerful subliminal cue was at work, not contained in the classical theory of stereopsis. We are working on models to explain this and other mysterious "unclassical" results.

Our recent models of the visual system depend on propagation of excitations in a two-dimensional network of neurons similar to those in the primary visual cortex. These models can detect a single moving dot in a field of thousands of fixed dots, in analogy with our ability to detect an artificial satellite moving against 3000 fixed stars in the night sky, for a signal/noise ratio of 1/3000 and seem to work well also for perception of shape, and depth. They are being developed using the largest unclassified Cray computer in addition to our own desktop machines and also being tested psychophysically. Simulations are essential since conventional mathematics are ill-suited to building a useful bridge between psychophysics and neurobiology.

http://mcb.berkeley.edu/faculty/NEU/glaserd.html

Category:

Science & Technology

Tags:

• CITRIS
• UCBerkeley
• Cognitive
• Computing
• Research
• Science
• Technology

[JOJOKYRA]

I easily can turn the direction of last, the "hardest" racetrack!?

[eswyatt]

Something about this bothers me. It seems it's getting something for nothing. All adding noise can do is correct for a threshold setting that, in the particular instance, is sub-optimal. If there is not the required sensitivity for sensing the differences in magnitude, then adding noise won't help the sensing device detect those changes.