This project implements a deep learning model that predicts classes which were unseen during training. This model learns the characteristics of the test classes using the semantics of the training classes. This project tests the use of attributes and word2vec for the semantic space.