Chaotic recurrent neural networks for brain modelling: A review

Area of Science:

• Computational Neuroscience
• Dynamical Systems Theory
• Artificial Intelligence

Background:

• The brain exhibits spontaneous, internally generated activity, often characterized by chaotic dynamics.
• Chaotic brain activity offers computational and behavioral advantages, including enhanced complexity and information processing.
• Traditional computational models often omit chaotic dynamics due to learning algorithm challenges.

Purpose of the Study:

• To review the computational benefits of chaos in brain activity.
• To examine chaotic recurrent neural networks (RNNs) and their training algorithms.
• To explore applications and limitations of chaotic RNNs in brain modeling.

Main Methods:

• Review of theoretical and experimental studies on brain chaotic dynamics.
• Analysis of algorithms for training chaotic recurrent neural networks (RNNs).
• Focus on reservoir computing paradigms and biologically plausible models.

Main Results:

• Chaos enhances network dynamics complexity and information storage/transfer.
• Algorithms have been developed to train chaotic RNNs, overcoming previous limitations.
• Chaotic RNNs show potential for brain modeling and broader applications.

Conclusions:

• Chaotic dynamics are integral to brain function, offering significant computational advantages.
• Advancements in training algorithms have made chaotic RNNs more viable for brain modeling.
• Chaotic RNNs represent a promising avenue for future neuroscience research and AI development.