Investigating interlimb generalisation of Bayesian sensorimotor learning

An emerging paradigm shift is currently underway in neuroscience involving the modelling of neural systems using the mathematical framework of Bayesian decision theory, and more significantly, treatment of the brain itself as a Bayesian machine. Recent work suggests that the brain represents probability distributions and performs Bayesian integration during sensorimotor learning, but the evidence remains inconclusive. In this study we seek to accrueadditional behabioural evidence for Bayesian sensorimotor learning. Using a novel variation of an interlimb generalisation paradigm involving a stochastic visuomotor perturbation, we tested the hypothesis that Bayesian sensorimotor learning transfers to the other limb, and relatedly, that the representation of this learned visuomotor perturbation is encoded in an extrinsic reference frame. Consistent with our hypothesis, we found that interlimb transfer of learning is nearly complete when the visuomotor perturbation is congruent in extrinsic coordinates and relatively incomplete when congruent in intrinsic coordinate. Interestingly, we also found that although the learned prior distribution transfers to the untrained limb, the likelihood is not optimally integrated by the untrained limb, indicating that the prior and likelihood are represented independently of one another. This study provides valuable information about the nature of the representations underlying Bayesian integration in sensorimotor learning and opens up a number of intriguing paths for future investigation.