• Robots

A computational model of action selection in the basal ganglia: thalamic and cortical interactions

M. D. Humphries & K. Gurney

This poster was presented at the FENS2000 Conference in Brighton, June 24th-28th 2000.

Reference: Humphries, M. D. & Gurney, K. (2000) A computational model of action selection in the basal ganglia: Thalamic and cortical interactions. European Journal of Neuroscience, 12, supp. 11, 260.10.

NOTE: The poster panel numbers are enclosed in brackets - clicking on them will open a panel window.

Introduction

This is the action selection problem: how does an animal choose which of a set of incompatible actions to do next? When deciding what to do an animal may have many possible actions to choose between but some of these actions may require the same motor resource and, therefore, could not be expressed simultaneously.
We recently proposed (Redgrave et al. 1999) that the vertebrate basal ganglia (BG) plays a pivotal role in the solution to the action selection problem. The BG are ideally placed to act as a central switching mechanism (Prescott et al. 1999) with massive input from the cortex and thalamus and outputs to brainstem and thalamic nuclei. Analysis and simulation of a computational model (Gurney et al. 1998a, 2000) [1] showed that intrinsic BG processing was capable of selecting and switching between actions [2]. The main shortcoming of this model was that it dealt with the BG in isolation and so did not show how the BG interacted with extrinsic brain structures. We present here a computational model that incorporates the BG into a thalamocortical loop.

Aims

Methods

The intrinsic BG model was incorporated into a thalamocortical loop comprising the ventrolateral (VL) thalamus and the primary motor cortex (M1) [3]; this was Model A. The addition of the TRN constituted Model B. The functional architecture of the TRN-VL connections [4] allow the TRN to act as a distal lateral inhibition mechanism for the VL.

Results[5]

Conclusion

The thalamocortical loop facilitates selection of low salience actions. The TRN’s role appears to be that of an automatic gain control (with respect to cortical input) or 'clean-up' circuit (with respect to basal ganglia output). The addition of anatomically constrained connections and structures to system level models of the basal ganglia consistently improves its ability to perform action selection, thereby supporting the computational hypothesis of action selection for these circuits.

References:

R. Albin, A. Young, J. Penney, Trends Neurosci. 12, 366 (1989)
L.L. Brown and F.R. Sharp, Brain Res. 686, 207 (1995)
G. Chevalier and J.M. Deniau, Trends Neurosci. 13, 277 (1990)
J. Cornwall, J. D. Cooper, O.T. Phillipson, Exp. Brain Res. 80, 157 (1990)
J.A. Gandia, S. De Las Heras, M. Garcia, J.M. Gimenez-Amaya, Brain Res. Bull. 32, 351 (1993)
R. W. Guillery, S. L. Feig, D.A. Lozsadi, Trends Neurosci. 21, 28 (1998)
K. Gurney, T. J. Prescott, P. Redgrave, Soc. Neuro. Abstr. 24, 640.13 (1998a)
K. Gurney, A. Prescott, P. Redgrave, AIVRU Technical Report 131 (1998b)
K. Gurney, A. Prescott, P. Redgrave, Biol. Cyber.(submitted)
J. E. Hoover and P. Strick, Science 259, 819 (1993)
E. G. Jones, The Thalamus (1985)
D. Pinault and M. Deschenes, Euro. J. Neurosci. 10, 3462 (1998)
T. J. Prescott, P. Redgrave, K. Gurney, Adaptive Behavior 7, 99 (1999)
P. Redgrave, T. J. Prescott, K. Gurney, Neuroscience 89, 1009 (1999)
I. Sumitomo and K. Iwama, Brain Res. 415, 389 (1987)