Topological exploration of artificial neuronal network dynamics

Area of Science:

• Neuroscience
• Computational Neuroscience
• Complex Systems

Background:

• Understanding neural network dynamics is crucial in neuroscience.
• Traditional methods like graph theory and statistics have limitations in characterizing complex network structures.
• A new approach is needed to analyze the spatiotemporal dynamics of interconnected neurons.

Purpose of the Study:

• To introduce a novel method using algebraic topology to classify neural network dynamics.
• To apply persistent homology for automatic classification of network activity based on topological features.
• To evaluate the efficacy of this topological approach in analyzing simulated neural networks.

Main Methods:

• Employed persistent homology, a tool from algebraic topology.
• Constructed topological spaces from various spike train distances.
• Computed spike train similarity measures and extracted topological features.
• Utilized a machine learning classifier trained on these topological features.

Main Results:

• The persistent homology method accurately classified simulated neural network dynamics into four regimes.
• A machine learning classifier trained on topological features generalized to unseen networks.
• Using features from multiple spike train distances systematically improved classification performance.

Conclusions:

• Algebraic topology offers a powerful new framework for analyzing neural network dynamics.
• Persistent homology provides a robust method for extracting meaningful topological features from neural activity.
• This approach enhances the ability to understand and predict complex neural network behaviors.