Brains, Meaning and Corpus Statistics by Tom M. Mitchell (cmu)

Tom Mitchell's web-page: http://www.cs.cmu.edu/~tom/
Based on “Predicting Human Brain Activity Associated with the Meanings of Nouns,”
Mitchell, Shinkareva, Carlson, Chang, Malave, Mason, & Just, Science, 2008.



1.   Can we train on word stimuli, then decode picture stimuli?
YES: We can train classifiers when presenting English words, then decode category of picture stimuli, or Portuguese words.
Therefore, the learned neural activation patterns must capture how the brain represents the meaning of input stimulus.
2.  Are representations similar across people?
Can we train classifier on data from a collection of people, then decode stimuli for a new person?
YES: We can train on one group of people, and classify fMRI images of new person.
Therefore, seek a theory of neural representations common to all of us (and of how we vary).
3.  Can we discover underlying principles of neural representations?
Idea: Predict neural activity from corpus statistics of
stimulus word [Mitchell
et
al.,
Science,
2008]




http://www.cs.cmu.edu/%7Etom/pubs/fMRI_public_May2009.pdf