A Neural Network Model of Sensorimotor Transformations in the Local Bending Reflex of the Medicinal Leech

SYNOPSIS. In response to a moderate mechanical stimulus, the leech withdraws by forming a U-shaped local bend beneath the stimulus. Interneurons in the local bending reflex receive multiple sensory inputs, and have outputs to many motor neurons, suggesting that stimulus location on the body is represented in a distributed fashion. However, it is possible that as yet unidentified interneurons dedicated to the detection of specific spatial locations are necessary for the full range of the local bending motor output. Using the backpropagation algorithm to optimize connections in a model of the reflex, we showed that the distributed processing model of sensorimotor integration was consistent with our knowledge of the connection strengths between identified neurons in the reflex. The model further showed that the spatial and temporal constraints on the linkage between sensory representations and motor output can be satisfied by the appropriate connectivity between layers of interneurons and motor neurons in the leech, as well as in other systems with similar circuitry.