Citation
R. Rustamov, L. Guibas. "Hyperalignment of Multi-Subject fMRI Data by Synchronized Projections". NIPS Workshop on Machine Learning and Interpretation in Neuroimaging (MLINI), 2013.

Abstract
Group analysis of fMRI data via multivariate pattern methods requires accurate alignments between neuronal activities of different subjects in order to attain competitive inter-subject classification rates. Hyperalignment, a recent technique pioneered by Haxby and collaborators, aligns the activations of different subjects by mapping them into a common abstract high-dimensional space. While hyperalignment is very successful in terms of classification performance, its anatomy free nature excludes the use of potentially helpful information inherent in anatomy. In this paper, we present a novel approach to hyperalignment that allows incorporating anatomical information in a non-trivial way. Experiments demonstrate the effectiveness of our approach over the original hyperalignment and several other natural alternatives.

Electronic downloads

• rg-hmsfdsp-2013.pdf · application/pdf · 244 kbytes

• HTML

  R. Rustamov, L. Guibas. <a
  href="http://robotics.eecs.berkeley.edu/pubs/16.html"
  >Hyperalignment of Multi-Subject fMRI Data by
  Synchronized Projections</a>, NIPS Workshop on Machine
  Learning and Interpretation in Neuroimaging (MLINI), 2013.

• Plain text

  R. Rustamov, L. Guibas. "Hyperalignment of
  Multi-Subject fMRI Data by Synchronized Projections".
  NIPS Workshop on Machine Learning and Interpretation in
  Neuroimaging (MLINI), 2013.

• BibTeX

  @inproceedings{RustamovGuibas13_HyperalignmentOfMultiSubjectFMRIDataBySynchronizedProjections,
      author = {R. Rustamov and L. Guibas},
      title = {Hyperalignment of Multi-Subject fMRI Data by
                Synchronized Projections},
      booktitle = {NIPS Workshop on Machine Learning and
                Interpretation in Neuroimaging (MLINI)},
      year = {2013},
      abstract = {Group analysis of fMRI data via multivariate
                pattern methods requires accurate alignments
                between neuronal activities of different subjects
                in order to attain competitive inter-subject
                classification rates. Hyperalignment, a recent
                technique pioneered by Haxby and collaborators,
                aligns the activations of different subjects by
                mapping them into a common abstract
                high-dimensional space. While hyperalignment is
                very successful in terms of classification
                performance, its anatomy free nature excludes
                the use of potentially helpful information
                inherent in anatomy. In this paper, we present a
                novel approach to hyperalignment that allows
                incorporating anatomical information in a
                non-trivial way. Experiments demonstrate the
                effectiveness of our approach over the original
                hyperalignment and several other natural
                alternatives.},
      URL = {http://robotics.eecs.berkeley.edu/pubs/16.html}
  }
  

Posted by Ehsan Elhamifar on 7 Jun 2014.
Groups: ehumans
For additional information, see the Publications FAQ or contact webmaster at robotics eecs berkeley edu.