[TNJC] Neuro Journal Club this week (21/10, 2pm, Tuesday)

[Ben Dongsung Huh]

Dear all, 

For tomorrow's TNJC, James Fitzgerald from Harvard University will present his work.

Title: The importance of natural sensory statistics for visual motion estimation

James E. Fitzgerald
Swartz Postdoctoral Fellow, Center for Brain Science, Harvard University

Abstract:

Many animals use visual signals to estimate motion. Canonical models suppose that animals estimate motion by computing pair correlations between spatiotemporally separated visual signals, but moving natural scenes contain many other visual signatures of motion. Here I'll argue that higher-order correlations are important components of the visual motion estimation strategies used by biology. I’ll begin by giving a general theoretical overview of how the visual signatures of motion depend on the statistics of visual stimuli. I’ll then present experimental evidence that motion-guided behaviors in humans, Drosophila, and zebrafish depend on higher-order stimulus correlations. Finally, I’ll present several simple models, constructed to architecturally resemble the Hassenstein-Reichardt Correlator, that predict key features of Drosophila’s higher-order motion response when optimized for motion estimation accuracy in natural environments. These results collectively support the viewpoint that biology makes purposeful use of the higher-order statistics of natural environments in order to improve the accuracy of its motion estimators.

See you then!

--------------​

Ben Dongsung Huh
Gatsby Computational Neuroscience Unit