Skip to main content

Winner-Take-All readout mechanisms in the Central Nervous System

Final Report Summary - WTAINCNS (Winner-Take-All readout mechanisms in the Central Nervous System)

Executive Summary:

Winner-Take-All readout mechanisms in the Central Nervous System.

Final Summary:

The scientific question and research approach.

What is the neural code? How is the information about the features of certain sensory stimulus or planned motor commands represented by the responses of large neuronal populations? And how does the central nervous system reads out this information in a manner that allows for accurate discrimination on brief biologically relevant timescales? These questions are at the heart of our proposed research.

We proposed to test the hypothesis that neuronal response latency may be the source of information that allows the central nervous system to yield fast and accurate decisions on brief timescales. We further suggested a latency-based competitive readout, termed the temporal Winner-Take-All (tWTA), where the tWTA estimates the external stimulus by the identity of the neuron that fired the first spike in the population.

To this end, we undertook a comprehensive multidisciplinary approach. In the first stage of our research we executed a rigorous analytical study, investigating our hypothesis using the framework of a simplified mathematical model. The analytical investigation enabled us to an analytical understanding of: what are the important parameters in the problem? and how these parameters affect the tWTA accuracy. Next, we extended our original research proposal to apply our theory to real neuronal data and test our hypothesis. We formed scientific collaborations with several experimental groups and studied latency coding in two sensory modalities and along different stages of the information processing pathway.

Main findings:

In our preliminary investigation we developed the essential theory of temporal-Winner-Take-All. We found that the temporal-Winner-Take-All can account for accurate discriminations between a small number of choices [1]. Next, we applied our resultant theory to study temporal-Winner-Take-All in the early visual system of the archer fish [2], and in the primary visual system of the monkey [4]. Our theory has also been applied to study the temporal-Winner-Take-All in the task of sound source localization in the auditory of the guinea pig [3].

Cells in the inferior colliculus (IC) code for the interaural time delay (ITD, a cue for sound source localization) of auditory stimuli by their rate of firing. Typically, ITD-coding IC cells in the guinea pig show a maximum firing rate in response to a 'preferred ITD'. Many cells in the guinea pig IC also show tuning of their first spike time latency to the ITD of the stimulus. Most cells have the shortest latency at their preferred ITD. We investigated the discrimination accuracy of a simple metric which estimates the stimulus ITD from the preferred ITD of the cell with the shortest latency, the temporal-Winner-Take-All. Surprisingly, despite being based upon only a single spike, we find that temporal-Winner-Take-All accuracy is comparable to the accuracy of a conventional rate-code metric, which takes into account the total number of spikes fired by the cell in the entire neural response to the auditory stimulus. The accuracy of the temporal-Winner-Take-All metric can be further increased by considering a generalized n- temporal-Winner-Take-All metric which estimates the auditory stimulus ITD as the preferred ITD of the cell which fired the first n spikes. Thus a metric based on spike latency may improve response speed at a small cost to the accuracy of the decision [3].

During the last stage of this project we turned to study the difficult problem of the computational implications of the inherent neuronal heterogeneity on the neural code and its accuracy. We addressed this issue using the framework of the neural responses of neurons in the IC of the ferret. We found that a simple readout that takes into account the specific structure of response diversity can overcome the drastically detrimental effect of empirically observed correlated noise fluctuations in the neuronal responses [5].


Beyond the basic scientific question of the neural code, obtaining a better understanding of the manner in which information is represented by the activity of large neuronal populations is central to our ability to build effective brain-machine interface. Our work has highlighted the computational advantages of encoding information by the neuronal response latency and has also shown that response latency in both the auditory and visual systems encodes considerable amount of information. We thus believe that incorporating ideas generated by our work will contribute greatly to the field of brain-machine interface.


Our findings have been presented in several international as well as national scientific meetings and have been published in peer reviewed journals and:

1. Shamir M. The Temporal Winner-Take-All Readout. PLoS Computational Biology 5(2): e1000286. (2009)
2. Vasserman G., Shamir M., Simon A. & Segev R. Coding 'What' and 'When' in the Archer Fish Retina. PLoS Computational Biology 6(11):e1000977 (2010).Oran
3. Zohar O, Shackleton TM, Nelken I, Palmer AR and Shamir M. First spike latency code for Interaural Time Delay discrimination in the guinea pig inferior colliculus. The Journal of Neuroscience 31(25):9192-204. (2011) PROJECT IN COLLABORATION WITH EUROPEAN LABORATORY.
4. Shriki O, Kohn A & Shamir M. Fast coding of orientation in the primary visual cortex. PLoS Computational Biology 8(6):e1002536 (2012).
5. Oran Zohar, Trevor M Shackleton, Alan R Palmer and Maoz Shamir. The effect of correlated neuronal firing and neuronal heterogeneity on population coding accuracy in guinea pig inferior colliculus. PLoS ONE 8(12): e81660 (2013).