Learning multi-whisker spatial-temporal cortical receptive fields from robotic whisker input

James Bednar, Stuart P. Wilson, Ben Mitchinson, Martin Pearson, Tony J. Prescott

Research output: Contribution to conferencePosterpeer-review


The arrangement of barrels in rodent primary somatosensory cortex (S1) forms a topographic map of the whiskers on the animal's snout. In layer 2/3 (L2/3) S1, there is evidence for a finer-scale map for the direction in which the individual whisker is deflected: Deflection of whisker A toward whisker B selectively activates neurons above barrel A that are located closest to barrel B (Andermann & Moore, Nat. Neurosci., 2006, 9(4): 543-551). More recently it has been shown that neurons in layer 5 (L5) S1 respond selectively to directions of global stimulus movement implied by the sequence in which an array of up to twenty-five whiskers is deflected (Jacob, Le Cam, Ego-Stengel & Schulz, Neuron, 2008, 60: 1112-1125). Here we explore whether Hebbian learning mechanisms can give rise to these phenomena, and present results from three computational models of the whisker barrel system.
First, we have shown previously that the L2/3 map organization can emerge in a developmental model (LISSOM) of S1 with lateral connectivity when trained on inputs from an array of simulated whiskers. We now show results from a simplified version of the model suggesting that the somatotopic whisker direction map arises when the combination of deflected whiskers is correlated with the direction in which the individual whisker is deflected during development. The improved model also predicts stronger connections between neurons that prefer similar deflection directions for different whiskers, emerging alongside the overall map structure, as suggested by recent findings in S1.
Second, we present evidence that multi-whisker inputs, capable of organizing receptive fields for single-whisker direction, also give rise to selectivity for multi-whisker deflection sequence in a self-organizing model extended to include a representation of L5 barrel cortex.
Third, results are presented from a model driven by simulated multi-whisker patterns, and by data recorded from robot-controlled collisions between an array of physical whiskers and a variety of tactile stimuli. This approach affords several testable predictions for how receptive fields for the spatial, temporal, and spatial-temporal features of tactile stimuli interact and are organized in primary sensory cortex.
Original languageEnglish
Publication statusPublished - Oct 2009
EventSociety for Neuroscience Annual Meeting 2009 - Chicago, United States
Duration: 17 Oct 200921 Oct 2009


ConferenceSociety for Neuroscience Annual Meeting 2009
Country/TerritoryUnited States


  • Barrel Cortex
  • Learning
  • Computational Model


Dive into the research topics of 'Learning multi-whisker spatial-temporal cortical receptive fields from robotic whisker input'. Together they form a unique fingerprint.

Cite this