Bayesian Learning for Neural Networks
Bayesian Learning for Neural Networks
Online Model Selection Based on the Variational Bayes
Neural Computation
| Hi-index | 0.00 |
We tried to reconstruct temporal movement information (position, velocity and acceleration) from single-trial brain activity measured using non-invasive methods. While human subjects performed wrist movement in eight directions, brain activity was measured by functional magnetic resonance imaging (fMRI) and magnetoencephalogram (MEG). To reconstruct the movement information, we used cortical currents estimated by hierarchical Bayesian method for each subject. Correlation coefficients between reconstructed position and actual position ranged from 0.45 to 0.6. Although accuracy of our method is inferior to those in a previous study, our method is based on cortical current that is tightly coupled with anatomical regions, and thus would be a useful tool in neuroscience if the accuracy could be improved.