Sparse synaptic connectivity is required for decorrelation and pattern separation in feedforward networks.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Neural Networks

Background:

• Pattern separation is a fundamental brain computation enabling distinct memory representations.
• Divergent feedforward networks are hypothesized to perform pattern separation via sparse coding.
• Spatial correlations in synaptic input and network connectivity may impair pattern separation performance.

Purpose of the Study:

• To investigate the structural and functional factors determining pattern separation in neural networks.
• To understand how synaptic connectivity influences the separation of spatially correlated input patterns.

Main Methods:

• Modeled the cerebellar input layer with spatially correlated input patterns.
• Systematically varied synaptic connectivity parameters.
• Quantified performance using the learning speed of a classifier trained on network outputs.

Main Results:

• Sparse synaptic connectivity is essential for separating spatially correlated input patterns across various network activity levels.
• Network expansion and the introduction of correlations were identified as major determinants of pattern separation.
• Sparse neuronal activity was found to be less critical than connectivity structure.

Conclusions:

• Sparse synaptic connectivity is a critical architectural feature for effective pattern separation, particularly in the presence of correlated inputs.
• The interplay between network expansion and correlation dynamics significantly impacts the brain's ability to differentiate similar patterns.
• Future research should focus on the role of network structure in pattern separation beyond simple sparse coding principles.