Closed emma-sjwang closed 6 years ago
Sorry for the delay in responding. This might depend on the assumptions that you make for how bag and instance labels are related. For example, a multi-class instance problem would imply a multi-label bag-level problem under the standard MIL assumption.
In any case, you could probably use a one-vs-rest classification scheme: treat each class as a separate binary problem, train a classifier for each, then recombine the outputs of the resulting classifiers to determine which of the multiple classes each bag or instance belongs to.
Thanks a lot.
I wonder how to realize multi label classification task using this binary classification MISVM.