Network hierarchy and pattern recovery in directed sparse Hopfield networks

Area of Science:

• Computational neuroscience
• Network science
• Systems biology

Background:

• Real-world networks are often directed, sparse, and hierarchical, featuring feedforward and feedback connections.
• A few key nodes frequently control the behavior of these complex systems.
• Understanding network structure is crucial for analyzing system dynamics.

Purpose of the Study:

• To investigate pattern dynamics in sparse, directed, Hopfield-like neural networks.
• To apply trophic analysis for characterizing network hierarchy and directionality.
• To identify key network properties influencing system control and performance.

Main Methods:

• Utilized trophic analysis to quantify node positions (trophic levels) and network directionality (trophic coherence).
• Studied the dynamics of pattern presentation and recovery in simulated neural networks.
• Performed numerical analysis to explore the impact of topological properties on network performance.

Main Results:

• Identified that a small subset of neurons with low trophic levels can control the system's state, even in recurrent networks.
• Demonstrated that pattern recovery performance significantly improves by optimizing trophic coherence and other topological features.
• Showcased the relevance of network structure in controlling system dynamics.

Conclusions:

• Sparse, hierarchical networks possess control mechanisms centered around low-trophic-level nodes.
• Network topology, particularly trophic coherence, is a critical factor for efficient pattern recovery.
• Findings offer insights into the structure of the animal brain and potential improvements for artificial neural networks and other complex systems.