Animals, and especially humans, can learn to flexibly adjust their movements to changing environments.
The neural principles underlying this remarkable capability are still not fully understood.
Among the most prominent brain regions controlling movement is primary motor cortex (M1).
Adapted motor behaviour can be related to a change in neural activity within this region. Yet, the
rules guiding this activity change, and thus behavioural adaptation, remain unclear. The overall aim
of this thesis is to investigate the learning process(es) governing the described change in activity in
M1 and, with that, the change in behaviour. Computational modelling is used to study three specific
aspects of learning:
1. What constrains learning to favour some neural activity patterns over others?
2. Can we identify where in a hierarchical pathway learning is happening?
3. How can sensory feedback guide the learning process?
We start by investigating what kind of biological constraints differentially affect learning of new
neural activity that either preserves coactivation patterns between neurons (within-manifold learning),
or requires learning of new coactivation patterns (outside-manifold learning). We propose a new
explanation - the learnability of feedback signals - for why within-manifold activity patterns can be
easier learned than outside-manifold activity patterns. In the second part we develop a hierarchical
model of the motor system to investigate whether we can derive where learning has happened from
only measuring neural activity. Lastly, we investigate how the brain could implement a biologically
plausible learning rule which allows it to correctly assign errors and update recurrent connectivity in
a goal-driven manner.
Overall, our work offers new perspectives on the role of M1 for motor learning and adaptation,
challenges current beliefs, and puts a focus on the role of feedback signals for local plasticity in M1.