Action Items
Issues
Active Learning
Experiments can use already labeled data
Simulate users with varying degrees and types of erroroneous behavior
Show that we can learn in the face of all user error types and establish measures of trust
Then involve real users
user trust
integrating mediawiki…. and using it to link to other people's datasets
system for contributing data sets….
use existing stand-off markup schemes (e.g., TEI)
allow for alternate tag sets with different granularities to enable people with varying degrees of ability
-
Computing EVSI quickly for an arbitrary joint probability model
-
Favorite Tasks
Classification tasks
UC Irvine set
contributed data sets
Back to top