More Insight Into Human Vision….From Machines

I love computers. My students are blissfully unaware of the challenges of typing a dissertation on a Smith-Corona portable electric typewriter or using a slide rule in a graduate statistics course. Still, I’d like to continue to think that the human brain can outperform computers, at least in some tasks.

One of our traditional advantages has been our ability to quickly categorize visual information. If usually takes us about 20 ms to decide whether an object is an animal or not. Now Thomas Poggio and his colleagues at MIT have designed a computer model that does a pretty good job on these rapid recognition tasks. [1]

The computer model also lends support to a hierarchical model of visual perception in rapid recognition tasks, in which simpler processing is built upon by more complex layers of processing. Eventually, one reaches a hypothetical “grandmother” cell that responds to the image of your grandma at the front door for a visit. Because the computer model does not use feedback loops, in which higher level processes revisit lower levels, while producing responses that were similar to human responses, Poggio and colleagues suspect that rapid recognition does not require feedback loops in people either.

The full-text of this article along with the stimuli used and other resources may be found on the PNAS website.

1.  Serre, T., Oliva, A., & Poggio, T. (2007). A feedforward architecture accounts for rapid categorization. Proceedings of the National Academy of Sciences, USA, 104(15), 6424-6429.