Concept classification with Bayesian multi-task learning

Van Gerven, Marcel; Simanova, Irina

Item

ITEM ACTIONSEXPORT

Add to Basket

Local TagsRelease HistoryDetailsSummary

Released

Conference Paper

Concept classification with Bayesian multi-task learning

MPS-Authors

/persons/resource/persons16260

Simanova, Irina
Neurobiology of Language Group, MPI for Psycholinguistics, Max Planck Society;

External Resource

No external resources are shared

Fulltext (restricted access)

There are currently no full texts shared for your IP range.

Fulltext (public)

VanGerven_Concept Classification with Bayesian Multi-task Learning.pdf
(Publisher version), 7MB

Supplementary Material (public)

There is no public supplementary material available

Citation

Van Gerven, M., & Simanova, I. (2010). Concept classification with Bayesian multi-task learning. In Proceedings of the NAACL HLT 2010 First Workshop on Computational Neurolinguistics (pp. 10-17). Los Angeles: Association for Computational Linguistics.

Cite as: https://hdl.handle.net/11858/00-001M-0000-0010-9095-B

Abstract

Multivariate analysis allows decoding of single trial data in individual subjects. Since different models are obtained for each subject it becomes hard to perform an analysis on the group level. We introduce a new algorithm for Bayesian multi-task learning which imposes a coupling between single-subject models. Using
the CMU fMRI dataset it is shown that the algorithm can be used for concept classification
based on the average activation of regions in the AAL atlas. Concepts which were most easily classified correspond to the categories shelter,manipulation and eating, which is in accordance with the literature. The multi-task learning algorithm is shown to find regions of interest that are common to all subjects which
therefore facilitates interpretation of the obtained
models.