Multi-class classificiation using active learning of fluorescent microscopy images based on their subcellular localization patterns.
The algorithm I used is a Query-by-committee based approach as explained by Nigam and McCallum. Ideally an active learning algorithm chooses only those points which will minimize the error upon querying their label. In QBC, this error is approximated by a group of features.