Motor programs: an artificial neural network approach

Abstract

It is commonly assumed that, during learning, the brain creates “motor programs” which store all the information essential to performing a motor skill. Yet there is still no consensus on what constitutes a motor program. In this study, a Multilayer Perceptron (MLP) network with one hidden layer, trained using the backpropagation rule, was used in an attempt to identify motor programs. Nine healthy subjects were asked to use their left hand to make fast and accurate movements in a tracking task of 75 identical steps, by either wrist flexion and extension, or the precision grip. The electromyogram (EMG) activity of 8 finger and hand muscles were simultaneously recorded by standard techniques. Onset timing of muscle activities were quantified from the digitized EMG signals, and were then used as the inputs to the MLP network. Reaction time was also measured, providing the desired output of the network. The trained network captured salient features of the relationship between EMG onset times and reaction time.