Cosyne 2007 Workshops
February 26-27, 2007
The Canyons, Utah
Speaker Name
Naftali Tishby (The Hebrew University)
Talk Title
Optimal adaptation and predictive information
Talk Abstract
We argue that an organism which maximizes the adaptive value of its actions given fixed resources must have internal representations of the outside world that are optimal in an abstract, information theoretic sense. This is true even if we don't know all the biological details that determine the metrics for costs and benefits. The resulting optimization principle "efficient representation of predictive information" includes as special cases problems in signal processing and learning that have clear connections to computations done by the brain. In an attempt to test this principle we suggest a new approach to the analysis of neural responses to sensory stimuli: Rather asking what features of the (past) stimulus triggered the neural response, we ask how much information the neural response provides about future stimuli. With carefully designed stimulus ensembles, one can measure directly this predictive information encoded by neural responses, independent of assumptions about the structure of the neural code. Using the motion sensitive neurons of the fly visual system as an example, we present preliminary experimental results that are in good agreement with predictions from our optimization principle.
Based partly on Joint work with Bill Bialek and Rob de Ruyter van Steveninck.
