Repository logo
 

Adaptive Prediction Error Coding in the Human Midbrain and Striatum Facilitates Behavioral Adaptation and Learning Efficiency.


Change log

Authors

Diederen, Kelly MJ 
Spencer, Tom 
Vestergaard, Martin D 
Fletcher, Paul C 

Abstract

Effective error-driven learning benefits from scaling of prediction errors to reward variability. Such behavioral adaptation may be facilitated by neurons coding prediction errors relative to the standard deviation (SD) of reward distributions. To investigate this hypothesis, we required participants to predict the magnitude of upcoming reward drawn from distributions with different SDs. After each prediction, participants received a reward, yielding trial-by-trial prediction errors. In line with the notion of adaptive coding, BOLD response slopes in the Substantia Nigra/Ventral Tegmental Area (SN/VTA) and ventral striatum were steeper for prediction errors occurring in distributions with smaller SDs. SN/VTA adaptation was not instantaneous but developed across trials. Adaptive prediction error coding was paralleled by behavioral adaptation, as reflected by SD-dependent changes in learning rate. Crucially, increased SN/VTA and ventral striatal adaptation was related to improved task performance. These results suggest that adaptive coding facilitates behavioral adaptation and supports efficient learning.

Description

Keywords

fMRI, normalization, performance, reward, standard deviation, Adaptation, Psychological, Corpus Striatum, Humans, Learning, Magnetic Resonance Imaging, Models, Psychological, Neuroimaging, Neurons, Psychomotor Performance, Reward, Substantia Nigra, Ventral Tegmental Area

Journal Title

Neuron

Conference Name

Journal ISSN

0896-6273
1097-4199

Volume Title

90

Publisher

Elsevier BV
Sponsorship
Wellcome Trust (093875/Z/10/Z)
Medical Research Council (MC_UU_12012/5)
Medical Research Council (MC_PC_12012)
This study was supported by the Wellcome Trust (W.S., P.C.F.), Bernard Wolfe Health Neuroscience Fund (P.C.F.) and the Niels Stensen Foundation (K.M.J.D.). We thank William Stauffer, Armin Lak and Joost Haarsma for useful discussions.