Representational learning by optimization of neural manifolds in an olfactory memory network

Area of Science:

• Neuroscience
• Computational Neuroscience
• Olfactory System Research

Background:

• Cognitive functions depend on internal representations organized by neural manifolds.
• Analyzing complex neural manifold geometry is challenging, especially without clear attractor dynamics.

Purpose of the Study:

• To investigate how olfactory discrimination training alters neural representations in the zebrafish pDp.
• To analyze the geometry of neural manifolds and its relation to odor discrimination behavior.

Main Methods:

• Trained juvenile and adult zebrafish in an odor discrimination task.
• Measured neuronal population activity in the telencephalic area pDp (zebrafish piriform cortex homolog).
• Analyzed representational manifolds using manifold capacity framework.

Main Results:

• No clear attractor dynamics were detected in the pDp.
• Olfactory training enhanced the separation of neural manifolds for task-relevant odors.
• Manifold capacity accurately predicted individual odor discrimination performance.

Conclusions:

• The pDp stores information in the geometry of neural manifolds, creating joint sensory-semantic maps.
• Geometric modifications of neural manifolds support the classification of sensory information.
• This geometric encoding may facilitate distributed learning processes in recurrent neural networks.