Redundancy in synaptic connections enables neurons to learn optimally.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Cortical microcircuits feature numerous multi-synapse neuron-to-neuron connections.
• The functional advantages of these redundant synaptic connections remain largely unexplored.

Purpose of the Study:

• To investigate the functional benefits of redundant synaptic connections in neural learning.
• To propose a computational framework linking synaptic plasticity and rewiring to learning.
• To model perceptual learning in the mammalian visual cortex.

Main Methods:

• Development of a simple dendritic neuron model to explore multisynaptic connections.
• Mathematical modeling of synaptic plasticity as particle filtering and wiring plasticity as resampling.
• Extension to a detailed single-neuron model of perceptual learning in the primary visual cortex.

Main Results:

• Redundant synaptic connections facilitate near-optimal learning when combined with synaptic rewiring.
• Synaptic plasticity with multisynaptic connections approximates particle filtering.
• Wiring plasticity implements the resampling step of particle filtering.
• The model accurately reproduces experimental observations in perceptual learning, including dendritic position-dependent plasticity and stimulus-selective synaptic organization.

Conclusions:

• Redundant synaptic connections are crucial for efficient learning in neural circuits.
• Synaptic plasticity and rewiring work together as a biological implementation of particle filtering for learning.
• The proposed framework offers a new conceptual understanding of synaptic plasticity and rewiring in the brain.