Artificial neural manifolds

Area of Science:

• Neuroscience
• Artificial Intelligence
• Materials Science

Background:

• The brain's rapid perception, prediction, and decision-making rely on coordinated neuronal activity forming a manifold structure.
• Most neuromorphic engineering approaches overlook this crucial neural manifold.
• Understanding and replicating this structure is key to advancing artificial intelligence.

Purpose of the Study:

• To construct an artificial neural manifold inspired by biological systems using Mott memristors.
• To develop a system capable of accurate and robust prediction with reduced sample requirements.
• To demonstrate the application of this artificial manifold in tasks like image perception and seizure prediction.

Main Methods:

• Construction of an artificial neuron circuit exhibiting biological neuron-like bell-shaped tuning curves.
• Convergence of large-scale neuronal firing into a compact, low-dimensional manifold structure.
• Application of the delay embedding theorem to establish a spatiotemporal information (STI) equation.
• Introduction of a memory factor to enhance the STI equation for improved prediction.

Main Results:

• The artificial neural manifold accurately represents information and enables prediction.
• The system achieves high accuracy and robustness in predictions, even with incomplete data.
• Reduced sample sizes are sufficient for effective learning and prediction.
• Successful perception of incomplete images and prediction of epileptic seizures were demonstrated.

Conclusions:

• The developed artificial neural manifold effectively mimics biological neural coordination for advanced cognitive tasks.
• This approach offers a promising pathway for creating more efficient and capable neuromorphic systems.
• The memory-enhanced STI equation significantly boosts prediction accuracy and robustness in complex scenarios.