Volitional Generation of Reproducible, Efficient Temporal Patterns

Area of Science:

• Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Biological brains exhibit remarkable energy efficiency compared to artificial intelligence (AI).
• Spike-based temporal codes are hypothesized to contribute to the brain's low energy consumption.
• The presence and learning dynamics of these codes beyond the sensory cortex remain largely unexplored.

Purpose of the Study:

• To investigate the implementation and evolution of energy-efficient temporal codes in the primary motor cortex (M1).
• To explore whether non-task-specific neurons also adopt energy-efficient strategies during learning.
• To understand the neural mechanisms underlying the refinement of these patterns through learning.

Main Methods:

• Development of a novel brain-machine interface (BMI) paradigm.
• Training two macaques to volitionally generate reproducible energy-efficient temporal patterns using M1 activity.
• Analysis of neuronal firing rates and temporal precision in task-related and non-task-related neurons.

Main Results:

• Macaques successfully learned to generate reproducible energy-efficient temporal patterns in M1 via the BMI.
• Neurons not directly involved in BMI control maintained energy efficiency, without increased excitability.
• Neuronal firing rates and temporal precision co-evolved during learning, indicating coordinated pattern refinement.

Conclusions:

• Energy-efficient temporal coding is implementable and learnable in the primary motor cortex (M1).
• Neural networks adapt cohesively to refine energy-efficient temporal patterns during learning.
• This study provides insights into the neural basis of efficient computation in the brain, relevant for AI development.