Energy Efficient Sparse Connectivity from Imbalanced Synaptic Plasticity Rules

Area of Science:

• Neuroscience
• Computational Neuroscience
• Evolutionary Biology

Background:

• Energy efficiency is a key constraint in brain evolution.
• Synaptic transmission is a major energy consumer in the brain.
• Sparse neural activity and connectivity may conserve energy.

Purpose of the Study:

• Investigate how sparse connectivity arises from synaptic plasticity rules.
• Explore the relationship between synaptic plasticity, information maximization, and energy efficiency.
• Provide a biophysically plausible mechanism for generating sparse synaptic configurations.

Main Methods:

• Studied a synaptic learning rule for excitatory synapses.
• Analyzed the impact of balanced versus imbalanced potentiation and depression.
• Investigated the connection between imbalanced plasticity and L1-norm regularization.
• Compared the proposed method with synapse pruning techniques.

Main Results:

• Information is maximized when synaptic potentiation and depression are balanced, yielding ~50% zero-weight synapses.
• Imbalancing plasticity towards depression increases zero-weight synapses without significant performance loss.
• Imbalanced plasticity acts as an L1-norm regularization, inducing sparseness.
• The proposed imbalanced plasticity is more efficient than synapse pruning.

Conclusions:

• Imbalanced synaptic plasticity provides a biophysically plausible mechanism for achieving sparse connectivity.
• This mechanism offers a novel explanation for the high prevalence of silent synapses in brain regions like the cerebellum.
• Sparse connectivity resulting from imbalanced plasticity contributes to neural energy efficiency and brain evolution.