Fifth International Conference on Cognitive and Neural Systems
May 30-June 2 - 2001 - Boston (Massachusetts, USA)
 
Rank Order Coding and Shunting Inhibition
 
(1)Simon J. Thorpe, (2)Laurent Perrinet & (1,3)Arnaud Delorme
 

(1)Centre de Recherche Cerveau & Cognition, Toulouse, France

(2)CERT-ONERA, Toulouse, France

(3)Computational Neurobiology Lab, Salk Institute, La Jolla, CA

thorpe@cerco.ups-tlse.fr, Laurent.Perrinet@cert.fr, arno@salk.edu

 
    Rank Order Coding uses the order in which neurons fire spikes to code information rather than their firing rates [1]. It has a number of interesting properties that include a remarkably high capacity (N neurons can transmit up to N! different orderings in a very short time window) and a built-in resistance to changes in overall input intensity and contrast. It has recently been shown that by using Rank Order Coding, the retina can transmit enough information to the visual cortex for image identification even
    when only a few percent of the retinal ganglion cells have fired a spike [2].
    Of course, even if information is contained in the order of firing, this information can only be used if the next neural processing stage can make use of the information. Here we describe a simple biologically plausible circuit that makes cortical neurons sensitive to the order of firing of their inputs. In the model, thalamic afferents make direct excitatory connections onto cortical pyramidal cells, but in addition, they have connections onto a pool of fast spiking inhibitory interneurons that produce shunting inhibition that affects the pyramidal cell population. This proposal fits with recent experimental data that have demonstrated a rapid increase in shunting inhibition in the visual cortex in response to a flashed stimulus[3]. The population of inhibitory interneurons effectively monitors the level of thalamic activity, with the amount of shunting inhibition increasing rapidly in response to a wave of thalamic inputs. As a result, consider what will happen when a group of afferents fires in a particular order. In the case of the first afferent to fire, shunting inhibition will be at a minimum, and thus the synapse will produce a maximum effect. Progressively, as the other inputs fire, the amount of shunting inhibition will build up so that the last inputs to fire will be strongly attenuated. Under such conditions, maximal post-synaptic activation will only be produced when the inputs fire in the order of their weights, with the inputs with the highest synaptic strengths firing first. By setting the post-synaptic spike threshold at an appropriate level, pyramidal neurons can be made sensitive to particular sequences of thalamic input. We will illustrate the model with simulations that show that such a mechanism can be used to generate a wide range of interesting neuronal properties that include orientation selectivity and sensitivity to particular acoustic patterns.
    1. Thorpe, S.J. and J. Gautrais, Rank Order Coding, in Computational Neuroscience: Trends in Research 1998, J. Bower, Editor.
    1998, Plenum Press: New York. p. 113-118.
    2. Van Rullen, R. and S.J. Thorpe, Rate coding versus temporal order coding: What the retinal ganglion cells tell the visual
    cortex. Neural Comput, 2001. In press.
    3. Borg-Graham, L.J., C. Monier, and Y. Fregnac, Visual input evokes transient and strong shunting inhibition in visual cortical
    neurons. Nature, 1998. 393(6683): p. 369-73.