Humans out-learning the machine: Support vector machines applied to fMRI of human motor learning

LaConte, S.; Chen, J.; Peltier, S.; Hu, X.

by S. LaConte, J. Chen, S. Peltier, X. Hu

Abstract:

This study examines whether learning by a subject leads to changes in the image data that are directly observable through statistical machine learning techniques. Five volunteers repeated a complex finger sequence over four runs, with behavioral learning (increased pace) compared to changes in performance of multivariate support vector machine regression (SVR) models. Our results demonstrated that SVR errors had a strong tendency to increase with time and had a strong positive correlation with the median number of button presses per run. We demonstrate that human learning is directly observable through increased prediction error in SVR models of fMRI data.

Reference:

abstract S. LaConte, J. Chen, S. Peltier, X. Hu. Humans out-learning the machine: Support vector machines applied to fMRI of human motor learning. In Proceedings 13th Scientific Meeting, International Society for Magnetic Resonance in Medicine, Miami, page 568, 2005. [bibtex]

Bibtex Entry:

@inproceedings{Miami568,
   Author = {LaConte, S. and Chen, J. and Peltier, S. and Hu, X.},
   Title ={Humans out-learning the machine: Support vector machines applied to f{M}{R}{I} of human motor learning},
   BookTitle = {Proceedings 13th Scientific Meeting, International Society for Magnetic Resonance in Medicine, Miami},
   Pages = {568},
   Abstract = {This study examines whether learning by a subject leads to changes in the image data that are directly observable through statistical machine learning techniques. Five volunteers repeated a complex finger sequence over four runs, with behavioral learning (increased pace) compared to changes in performance of multivariate support vector machine regression (SVR) models. Our results demonstrated that SVR errors had a strong tendency to increase with time and had a strong positive correlation with the median number of button presses per run. We demonstrate that human learning is directly observable through increased prediction error in SVR models of fMRI data.},
 Keywords = {Miami568},
   Year = {2005} }