Departmental Colloquium: David Burr, Pisa Vision Lab

Type: 
Colloquia
Audience: 
CEU Community Only
Building: 
Oktober 6 u. 7
Room: 
101
Wednesday, October 26, 2016 - 5:00pm
Add to Calendar
Date: 
Wednesday, October 26, 2016 - 5:00pm to 6:30pm

Positive and negative serial-dependencies in face perception

David Burr Department of Neuroscience, University of Florence, Italy

 

Perception is driven not only by the stimuli currently impinging on our senses, but also depends on the immediate past history. I will talk about these serial dependencies, showing how they are the product of efficient mechanisms exploiting temporal redundancies in natural scenes, and may be critical for understanding perception. Serial dependencies can be negative – such as visual aftereffects – where viewing motion in one direction causes stationary stimuli to appear to move in the other; or they can be positive – such as “priming” – where viewing a stimulus distorts subsequent stimuli in the same direction. Many factors influence whether the dependencies are positive or negative, including the strength and salience of the priming stimulus. However, if the serial dependencies reflect efficient processing, they should also depend on the attribute being tested. Negative aftereffects optimize sensitivity to change, while positive serial dependencies integrate successive image views, improving signal-to-noise ratios. On this logic, positive dependencies should occur for stable attributes – such as identity and gender – and negative dependencies for changeable attributes – such as expression. Indeed, when we asked subjects to judge both the expression and gender of a sequence of faces, we found strong and consistent positive serial dependencies for gender, but negative serial dependency for expression. These results show that both positive and negative serial dependencies can operate at the same time, on the same stimuli, depending on the attribute being judged, pointing to very flexible and sophisticated optimization of past information. The results were well modelled by a Kalman filter type model, showing that the strategies led to improvement in efficiencies.