Learning cortical representations from multiple whisker inputs

Abstract

Rats' whiskers convey tactile information to thesomatosensory cortex, where layer 4 neurons areclustered into barrels, each responding primarily toinput from one principal whisker (PW). The spatialarrangement of the barrels reflects the spatialarrangement of the whiskers on the animal's snout,thus representing the whiskers in a somatotopicmap. Within a barrel, neurons are selective for thedirection in which the PW is deflected, and acrosslayer 2/3 directions may be organized into apinwheel map such that deflection of whisker Atowards whisker B activates barrel field A neuronslocated closest to barrel field B (Figure 1c). Morerecently layer 5 neurons have been found to beselective for the direction in which waves ofsequential deflections are applied across multiplewhiskers, although the potential spatialorganization of a map for these stimuli has not yetbeen determined. In previous work we have shown howsingle whisker direction maps can emerge from aLISSOM (laterally interconnected synergeticallyself-organizing map) model of layer 2/3 barrelcortex, when the directions of waves of multiwhisker input are correlated with the directions ofthe individual whiskers. Here we investigate theemergence and organization of multi whiskerrepresentations in an additional sheet of layer 5neurons. Self organization of this system is drivenby signals measured from an array of simulatedwhiskers, however, work is currently in progress togenerate training data from an array of physicalcomposite glass fiber whiskers (Figure 1b) mountedon an XY translation table. The hardware basedapproach allows us to investigate how corticalrepresentations for temporal features such asstimulus onset/offset, velocity and frequency mightbe integrated with those for the spatial componentsof whisker stimuli, and should enable us to predictreceptive field properties of layer 5 cells that maybe measurable in future in vivo experiments.