Synaptic Pruning and Firing Rate Analysis

Overview

This project investigates the effects of synaptic pruning on neural network dynamics using NEST Simulator. We analyze how different pruning strategies impact network activity, particularly focusing on excitatory-inhibitory interactions and firing rates under varying thalamic input frequencies.

Pruning Strategies

Two pruning methods are implemented:

• Distance-based pruning: Removes synapses exceeding a set Euclidean distance threshold.
• Weight-based pruning: Removes synapses with an effective weight (computed as ( u \times x )) below a given threshold.

A baseline model with no pruning is used for comparison.

Simulation Setup

1. Network Initialization

   • Creates excitatory (layer_e) and inhibitory (layer_i) neuron populations.
   • Establishes synaptic connections based on predefined connectivity rules.

2. Pruning Application

   • Pruning strategies are applied independently to ensure results are not influenced by prior pruning steps.

3. Thalamic Input Simulation

   • The network is stimulated with varying thalamic input rates to assess response dynamics.

4. Recording and Analysis

   • Measures synaptic variables (u, x) and firing rates across different conditions.
   • Results are visualized to compare pruning effects.

Results

• Pruning leads to increased excitability: Pruned networks show higher firing rates compared to the baseline.
• Weight pruning has a stronger effect: It causes greater deviations from the unpruned model compared to distance pruning.
• Activity stabilizes at higher input frequencies: The difference between pruned and unpruned networks diminishes as input rates increase.

Plots and Analysis

1. Firing Rate Differences

Plots illustrate how pruning alters firing rates relative to the baseline, revealing stronger deviations at moderate input frequencies.

2. Synaptic Variable Evolution

Tracking u (utilization) and x (available resources) over time highlights distinct recovery and depletion patterns under different pruning conditions.

3. Input-Response Relationship

Comparisons of firing rates as a function of thalamic input provide insight into network adaptation and stability.

4. Interactive Plots using Plotly

Comparison of the firing rates of the networks with and without pruning giving two constant DC currents as input.

Future Improvements

• Implement additional pruning strategies (e.g., activity-based pruning).
• Explore network plasticity and long-term stability.
• Optimize computational efficiency for large-scale simulations.