Neural manifold under plasticity in a goal driven learning behaviour

Author summary It has been suggested that the coordinated activation of neurons might play an important role for movement execution. Whether such activity patterns are fixed or flexibly relearned remains matter of debate. It has been shown that monkeys can learn within minutes to adjust their neural activity, as long as they use the initial set of activity patterns. In contrast, monkeys needed several days and a sequential training procedure to learn completely new patterns. Here, we developed a computational model to investigate which biological features might lead to these experimental observations. Learning in our model is implemented through weight changes between neurons in a recurrently connected network. In order for these weight changes to improve the produced behaviour, an error signal is required which tells each neuron whether it should increase or decrease its activity in order to produce a movement closer to the target movement. We found that learning such an error signal is possible only in the first experimental condition, where monkeys needed to adapt their neural activity using already existing activity patterns. The learning of this error signal therefore poses a major constraint on what type of changes in neural activity can and can not be learned. Neural activity is often low dimensional and dominated by only a few prominent neural covariation patterns. It has been hypothesised that these covariation patterns could form the building blocks used for fast and flexible motor control. Supporting this idea, recent experiments have shown that monkeys can learn to adapt their neural activity in motor cortex on a timescale of minutes, given that the change lies within the original low-dimensional subspace, also called neural manifold. However, the neural mechanism underlying this within-manifold adaptation remains unknown. Here, we show in a computational model that modification of recurrent weights, driven by a learned feedback signal, can account for the observed behavioural difference between within- and outside-manifold learning. Our findings give a new perspective, showing that recurrent weight changes do not necessarily lead to change in the neural manifold. On the contrary, successful learning is naturally constrained to a common subspace.

Automated summary

The study investigates the role of neural activity patterns in movement execution, showing that monkeys can quickly adapt their neural activity but learning an error signal is a major constraint for new patterns. The findings suggest that successful learning is naturally constrained to a common subspace, providing a new perspective on motor control mechanisms.