Neural constraints on learning.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Motor Control

Background:

• Learning new skills, motor, sensory, or cognitive, involves neural networks generating novel activity patterns.
• The ease of learning some behaviors suggests that certain neural activity patterns might be easier to generate than others, potentially due to network constraints.

Purpose of the Study:

• To investigate if existing neural network structures constrain the generation of new neural activity patterns.
• To identify principles governing these constraints and their impact on learning.

Main Methods:

• Utilized a closed-loop intracortical brain-computer interface (BCI) learning paradigm with Rhesus macaques.
• Monkeys controlled a computer cursor by modulating neural activity in the primary motor cortex, with the neural activity-to-cursor velocity mapping manipulated by the BCI.

Main Results:

• Neural population activity naturally resides within a low-dimensional subspace termed the intrinsic manifold, reflecting network circuitry constraints.
• Monkeys readily learned cursor control using neural activity within the intrinsic manifold.
• Learning was significantly impaired when attempting to use neural activity patterns outside the intrinsic manifold.

Conclusions:

• The intrinsic structure of neural networks shapes and constrains learning.
• Generating neural activity patterns inconsistent with the network's intrinsic structure is difficult on short timescales (hours).
• These findings provide a network-level explanation for why related skills are learned more easily.