Microstimulation of a neural-network model for visually guided saccades

Area of Science:

• Neuroscience
• Computational Neuroscience
• Ophthalmology

Background:

• Eye movements are crucial for visual perception and are controlled by complex neural circuits.
• Previous models struggled to explain eye movement coordination using purely retinal or head-centered frames of reference.

Purpose of the Study:

• To investigate the coordinate coding of eye movements using a trained neural network model.
• To compare the effects of microstimulation in the model with those observed in biological systems.

Main Methods:

• A neural network was trained to map retinal and eye position inputs to head-centered coordinates.
• Microstimulation was applied to single units in the middle layer of the trained network.
• The resulting simulated eye movements (saccades) were analyzed.

Main Results:

• Microstimulation of the neural network produced saccades similar to those observed in brain stimulation experiments.
• The model's middle-layer unit activity represented desired eye positions in head-centered coordinates.
• Stimulation did not yield saccades predicted by classical head-centered coding, suggesting a distributed coding mechanism.

Conclusions:

• The neural network model successfully replicates complex eye movement generation observed in the brain.
• Eye movement spatial coding appears to be distributed across populations of neurons rather than localized to single cells.
• This finding challenges classical models and offers insights into the neural basis of spatial representation.