A Spiking Neural Network Model of Rodent Head Direction Calibrated With Landmark Free Learning.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Head direction cells are crucial for spatial orientation.
• Maintaining accurate head direction relies on integrating self-motion (idiothetic) and environmental (allothetic) cues.
• In environments lacking visual landmarks or in darkness, idiothetic cues alone lead to heading estimate drift.

Purpose of the Study:

• To investigate how animals learn to associate visual scenes with head direction without explicit landmark extraction.
• To develop and test a spiking neural network (SNN) model for head direction.
• To demonstrate that visual learning algorithms can correct drift in idiothetic-based head direction estimates.

Main Methods:

• Utilized discriminative and generative visual processing methods.
• Developed a spiking continuous attractor model (SNN) for head direction.
• Trained model-free visual learning algorithms on SNN-generated head angles.
• Validated the model using cue rotation experiments.

Main Results:

• Both discriminative and generative visual methods effectively provided corrective signals for head direction.
• The SNN model, when driven by idiothetic input, exhibited drift.
• Visual learning algorithms successfully corrected SNN drift, even with limited training data.
• Model predictions accurately reproduced experimental findings in cue rotation paradigms.

Conclusions:

• Visual information is essential for stabilizing head direction estimates, compensating for idiothetic drift.
• Model-free visual learning can learn associations with head direction without explicit landmark identification.
• The developed SNN model and learning algorithms provide a viable framework for understanding head direction stabilization in navigating animals.