Learning accurate path integration in ring attractor models of the head direction system

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Ring attractor models are supported by experiments for angular path integration.
• Head direction circuits require precise connectivity for accurate integration.
• The mechanism for achieving this precise tuning in neural circuits remains largely unknown.

Purpose of the Study:

• To propose a biologically plausible network model for learning synaptic efficacies in head direction circuits.
• To investigate how supervisory cues can guide the development of path integration capabilities.
• To apply the model to the *Drosophila* head direction system and compare its learned connectivity to experimental data.

Main Methods:

• Development of a local, biologically plausible learning rule adjusting synaptic efficacies.
• Application of the learning rule to a model of the *Drosophila* head direction system.
• Simulation of network behavior under optogenetic stimulation and comparison with experimental findings in flies and rodents.

Main Results:

• The model successfully learns to path-integrate angular information accurately.
• The learned synaptic connectivity closely resembles experimentally observed patterns.
• The mature network functions as a quasi-continuous attractor, reproducing experimental results on heading representation and remapping.

Conclusions:

• Self-supervised learning during a developmental phase is crucial for accurate path integration.
• The proposed framework offers a general mechanism for learning path integration, even in non-ring architectures.
• The model provides insights into the developmental processes underlying spatial navigation circuits.