I am working on finding points which are at boundary of different classes. In other words finding points on which a classifier would be most confused or uncertain about. For a setting like multi label classification is there a novel method to get these points so that I use them to do Active learning.
There are a couple of query algorithms to do so, see e.g. QBC algorithm. The basic idea here is to train a set of diverse and reliable learners/models on the currently labeled dataset. Then you use these models to predict the class of each unlabelled datapoint. For a certain datapoint, If all the learners predict the same class, you conclude that you have enough information about it. Otherwise, you ask the user to label it.
Another example is uncertainty sampling. However, both these algorithms ignore the true distribution of the data. Hence, some improvements have been suggested, see e.g. QBF.
A good reference about different algorithms and query strategies is this literature survey by Burr Settles.