@misc{Galetzka2018,
  type      = {Master Thesis},
  author    = {Galetzka, Cedric},
  title     = {Reward and prediction errors in Bayesian sensorimotor control},
  doi       = {10.25932/publishup-50350},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus4-503507},
  school      = {Universit{\"a}t Potsdam},
  pages     = {53},
  year      = {2018},
  abstract  = {Midbrain dopamine neurons invigorate responses by signaling opportunity costs (tonic dopamine) and promote associative learning by encoding a reward prediction error signal (phasic dopamine). Recent studies on Bayesian sensorimotor control have implicated midbrain dopamine concentration in the integration of prior knowledge and current sensory information. The present behavioral study addressed the contributions of tonic and phasic dopamine in a Bayesian decision-making task by alternating reward magnitude and inferring reward prediction errors. Twenty-four participants were asked to indicate the position of a hidden target stimulus under varying prior and likelihood uncertainty. Trial-by-trial rewards were allocated based on performance and two different reward maxima. Overall, participants' behavior agreed with Bayesian decision theory, but indicated excessive reliance on likelihood information. These results thus oppose accounts of statistically optimal integration in sensorimotor control, and suggest that the sensorimotor system is subject to additional decision heuristics. Moreover, higher reward magnitude was not observed to induce enhanced response vigor, and was associated with less Bayes-like integration. In addition, the weighting of prior knowledge and current sensory information proceeded independently of reward prediction errors. Taken together, these findings suggest that the process of combining prior and likelihood uncertainties in sensorimotor control is largely robust to variations in reward.},
  language  = {en}
}