Active Learning
Active Learning algorithm can attain accuracy with fewer training labels if it carefully selects the data from which it learns. An active learner may submit queries, typically as unlabeled data instances to be labeled by the user. In situations where unlabeled data is abundant or easily obtained, and manual labeling is difficult, expensive and time consuming, learning algorithm can actively query the user for labels.
Goals:
Generate a robust classifier, without having to mark up and source the learner with more data than needed. It works towards keeping the human labeling effort to minimum, only requiring guidance where the training utility of the outcome of such a query is high.
BrandIdea’s Implementation:
Product recognition & tagging (image analysis)