shadow_tr
Quick Links
 
• 
 
 
• 
 

 

Events


2014-04-15 - Kathleen Hoffman
 
2014-10-14 - Howard Elman
 

News


No pages meet the criteria

2014-03-20 - Julijana Gjorgjieva


Time/Place:

Thursday, March 20, 2014
4:00pm
Jones Room 204
Tulane University (Uptown)

Refreshments will be served


Speaker:

Julijana Gjorgjieva, Center for Brain Science, Harvard University


Title:

Optimal Sensory Coding by Populations of ON and OFF Neurons


Abstract:

Sensory responses of a population of neurons generally split in two types: ON cells that respond to stimulus onset and OFF cells that respond to stimulus offset. The ON-OFF dichotomy has been observed in different sensory modalities and species, including the vertebrate retina, invertebrate motion vision, the auditory and the olfactory systems. For example, ON ganglion cells in the retina respond when the light intensity increases, while OFF ganglion cells respond when the light intensity decreases. It has been proposed that the ON-OFF splitting has emerged to encode signal increments and decrements in an energy-efficient way, i.e. by using lower firing rates.

I will discuss the conditions under which sensory coding by a mixture of ON and OFF cells is more efficient than coding by a population of only one cell type (ON or OFF) using two measures of efficiency: the mutual information between stimulus and response, and the accuracy of a linear decoder. Surprisingly, our results show that the information transmitted by a neuronal population is independent of the way the population splits into ON and OFF cells.  However, an equal mixture of ON and OFF subpopulations is the most efficient in the sense that it uses fewest spikes. In addition to computing information, if we ask for downstream neurons to linearly decode the stimulus, the ON-OFF system can achieve a better stimulus decoding. We can relate the statistics of natural stimuli to the optimal ON-OFF ratio in a population of neurons. Given that dark edges predominate in natural images, or theory predicts more OFF than ON cells in the population, in agreement with experiments. We compare our predictions for the optimal distribution of thresholds to those of experimentally recorded ganglion cells in the salamander retina.

Center for Computational Science, Stanley Thomas Hall 402, New Orleans, LA 70118 ccs@tulane.edu