Complex dynamics and the structure of small neural networks.

Area of Science:

• Computational neuroscience
• Complex systems dynamics
• Nonlinear dynamics

Background:

• Neural networks exhibit complex behaviors beyond simple signal processing.
• Understanding the discrete-time dynamics of small networks is crucial for neuroscience and computational modeling.
• Bifurcation scenarios in neural network models can lead to a rich variety of dynamical states.

Purpose of the Study:

• To empirically investigate the discrete-time dynamics of small neural networks.
• To identify and characterize non-trivial bifurcation scenarios and their resulting dynamical properties.
• To define equivalence classes of networks based on dynamical capacities and explore symmetry-generated variations.

Main Methods:

• Empirical analysis of discrete-time dynamics in two- and three-neuron networks.
• Identification of periodic, quasi-periodic, and chaotic attractors.
• Definition of network equivalence classes based on parametrized dynamical systems.
• Analysis of phase space and parameter space symmetries.

Main Results:

• Observed co-existing periodic, quasi-periodic, and chaotic attractors for identical parameter sets in small neural networks.
• Defined an equivalence class for networks with identical dynamical capacities.
• Demonstrated how combined symmetries in phase and parameter spaces generate distinct network representatives.
• Suggested connectivity conditions ensuring complex dynamics within equivalence classes.

Conclusions:

• Small neural networks can exhibit complex dynamics, including co-existing attractors, even with simple structures.
• Network equivalence classes and their symmetries provide a framework for understanding diverse dynamical behaviors.
• Specific connectivity structures are critical for guaranteeing complex dynamics in these systems.