Neural manifold under plasticity in a goal driven learning behaviour.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Motor Control

Background:

• Neural activity is typically low-dimensional, characterized by prominent covariation patterns.
• These patterns are hypothesized to be fundamental for rapid and adaptable motor control.
• Monkeys demonstrate rapid adaptation of motor cortex activity within a low-dimensional subspace (neural manifold).

Purpose of the Study:

• To elucidate the neural mechanisms behind within-manifold adaptation in motor cortex.
• To explain the behavioral differences observed in learning within versus outside the neural manifold.
• To investigate how recurrent weight modifications influence neural manifolds during learning.

Main Methods:

• Development of a computational model simulating neural activity and motor control.
• Incorporation of recurrent weight modification driven by a learned feedback signal.
• Analysis of model behavior to differentiate between within- and outside-manifold learning.

Main Results:

• Recurrent weight modification, driven by feedback, successfully models within-manifold adaptation.
• The model demonstrates that learning can occur without altering the underlying neural manifold.
• Behavioral differences between within- and outside-manifold learning are explained by the model.

Conclusions:

• Learned feedback signals driving recurrent weight changes can explain rapid motor adaptation within a neural manifold.
• Recurrent weight modifications do not necessarily change the neural manifold; learning is constrained to it.
• This provides a mechanism for flexible motor control that leverages, rather than alters, intrinsic neural structure.