One nose but two nostrils: Learn to align with sparse connections between two olfactory cortices

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Inter-hemispheric integration of neural representations is crucial in neuroscience.
• Odor responses in cortical neurons show high correlation between nostrils, suggesting structured connections.
• The precise mechanism for this bilateral alignment remains unclear.

Purpose of the Study:

• To investigate how continuous odor exposure shapes inter-hemispheric projections.
• To model this process using online learning with a local Hebbian rule.
• To compare Hebbian learning with global stochastic-gradient-descent (SGD) for artificial neural networks.

Main Methods:

• Modeled inter-hemispheric projection shaping as online learning with a local Hebbian rule.
• Analyzed the trade-off between speed and accuracy in Hebbian learning.
• Investigated the relationship between cortical neuron count and projection density.
• Compared Hebbian learning with SGD in artificial neural networks.

Main Results:

• Hebbian learning with sparse connections successfully achieves bilateral alignment.
• A linear trade-off between learning speed and accuracy was observed.
• An inverse scaling relationship exists: more cortical neurons permit sparser inter-hemispheric projections for alignment.
• SGD achieved similar alignment accuracy with sparser connectivity, following the same scaling relationship.

Conclusions:

• Local Hebbian learning provides an effective mechanism for bilateral neural alignment.
• The observed inverse scaling relationship offers insights into efficient neural architecture.
• Similar performance between Hebbian and SGD learning stems from aligned update vectors, suggesting potential for sparse, local learning algorithms.