Future Directions
n Discovering Hidden Variables
n Can we use Active Learning to guide us?
n Coping with Missing or Unlabeled Data
n Use unlabeled pool to improve SVM performance
n Allow incorporation of partial experimental results
n Different action costs
n Performing some experiments more costly than others
n Extra cost to observe some domain variables
n E.g., should we perform NMR scan on this patient or not?
n Scaling up to even larger domains
n Images databases with millions of images