The oculomotor integrator: testing of a neural network model

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• The neural integrator in the caudal pons is crucial for the vestibulo-ocular reflex, converting eye-velocity to eye-position commands.
• Previous models simulated this function but employed unphysiological learning algorithms.

Purpose of the Study:

• To develop a new neural network model of the neural integrator that utilizes a physiologically plausible, Hebbian-like learning algorithm.
• To investigate the role of commissural connections in neural integration and compare model behavior to experimental data.

Main Methods:

• A recurrently connected neural network with a local, Hebbian-like learning rule was developed to simulate neural integration.
• The model learned to integrate semicircular canal signals and compensate for orbital mechanics by minimizing retinal slip error.
• Electrical perturbations and lesions were performed on awake rhesus monkeys to study the neural integrator's function in vivo.

Main Results:

• The model successfully learned to perform neural integration and compensate for orbital mechanics using a physiologically plausible learning rule.
• The network demonstrated recovery from lesions and adaptive gain changes under simulated abnormal visual-vestibular interactions.
• Experimental perturbations in monkeys, particularly involving midline pontine stimulation and lesions, showed results consistent with the model's behavior, supporting the role of an inhibitory commissure.

Conclusions:

• A novel, physiologically plausible neural network model effectively simulates the neural integrator's function.
• Commissural connections, particularly through lateral inhibition, are essential for the neural integrator's operation.
• The findings support the hypothesis that positive feedback via an inhibitory commissure underlies neural integration in the brainstem.