Correlated neural variability in persistent state networks.

Area of Science:

• Computational neuroscience
• Systems neuroscience

Background:

• Persistent neural activity underlies short-term memory but is vulnerable to drift from spiking variability.
• Existing models often assume independent neural inputs, contradicting experimental data showing correlated variability.

Purpose of the Study:

• To investigate how correlated variability in neural networks affects the stability of persistent activity and memory task performance.
• To determine the optimal structure of correlated variability for robust short-term memory.

Main Methods:

• Analysis of stochastic attractor dynamics in mutually inhibitory spiking neural networks.
• Utilized a reduced firing rate model to analyze persistent states and stochastic drift.
• Simulated performance on a two-interval, delayed response discrimination task.

Main Results:

• Globally distributed correlated variability across neural populations reduced variability in persistent activity compared to local correlations.
• Correlated inputs, when distributed globally, induced fluctuations orthogonal to the attractor, minimizing stochastic drift.
• Global distribution of correlated variability improved network performance on the memory discrimination task.

Conclusions:

• The correlation structure of neural input fluctuations is critical for maintaining stable persistent activity.
• Distributing correlated variability globally, rather than locally, enhances the robustness and performance of neural networks supporting short-term memory.