Neurophysiological Correlates of Probabilistic Reward-Based Learning: Using Decoding Approach on MEG Data

Prediction error and volatility estimate are important concepts in the predictive coding theory. In the present study, we derive the values of prediction error and volatility estimate from a hierarchical Bayesian model - Hierarchical Gaussian Filter. Using support vector machine (SVM) method, we predict the values of prediction error and volatility estimate from brain activity measured by magnetoencephalography (MEG). Our findings suggest that these computational values are indeed represented in the neural data, supporting the neural basis of predictive coding mechanisms.