This document summarizes a research paper titled "Fast Image Tagging" presented at ICML2013. The paper proposes a new method called FastTag that can quickly and accurately tag images with relevant keywords. FastTag learns a mapping from image features to a completed tag set by simultaneously training two classifiers - one to predict the complete tag set from images, and another to enrich existing sparse tags. It uses a marginalized blank-out regularization technique to guide the learning without needing corrupted training data. Experiments show FastTag achieves state-of-the-art accuracy on par with previous best methods but with faster training and testing times.