Attractor and integrator networks in the brain.

Area of Science:

• Computational neuroscience
• Cognitive neuroscience
• Theoretical neuroscience

Background:

• The brain's ability to maintain persistent activity is crucial for cognitive functions like working memory.
• Understanding the neural mechanisms underlying these persistent states has been a long-standing challenge.
• Attractor neural network models offer a promising framework for explaining these complex dynamics.

Purpose of the Study:

• To review the success of attractor neural network models in explaining brain functions.
• To explore the mechanisms by which neural units generate long-timescale dynamics.
• To discuss the computational applications of attractor dynamics in the brain.

Main Methods:

• Review of theoretical attractor neural network models.
• Analysis of collective dynamics from simple and forgetful units.
• Identification of continuous-attractor dynamics in brain systems.

Main Results:

• Attractor models successfully describe persistent activity, error correction, and cue integration in the brain.
• Simple neural units can collectively generate complex, long-timescale dynamics.
• Continuous-attractor dynamics have been rigorously identified in specific brain systems.

Conclusions:

• The brain demonstrably constructs and utilizes attractor systems for computation.
• Modular attractors can be reused and recombined for multiple functions, balancing robustness, capacity, and flexibility.
• Recent theoretical advances provide insights into overcoming trade-offs in neural representation.